Hedge Funds - Alternative Investment Strategies and...
Transcript of Hedge Funds - Alternative Investment Strategies and...
Hedge Funds
Alternative Investment Strategies andPortfolio Models
Wolfgang Mader
Erstgutachter: Prof. Dr. Manfred Steiner
Zweitgutachter: Prof. Dr. Gunter Bamberg
Vorsitzender der mundlichen Prufung: Prof. Dr. Hans Ulrich Buhl
Datum der mundlichen Prufung: 22. Juli 2005
Dissertation zur Erlangung des Grades eines Doktors der
Wirtschaftswissenschaften (Dr. rer. pol.) durch die Wirtschaftswissenschaftliche
Fakultat der Universitat Augsburg
Contents
List of Figures VIII
List of Tables XII
List of Abbreviations XV
List of Symbols XIX
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objectives and Structure . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Hedge Funds 7
2.1 Hedge Funds Defined . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Hedge Funds in the Alternative Investment Universe . . . . . . . . . 10
2.3 Key Characteristics of Hedge Funds . . . . . . . . . . . . . . . . . . . 12
2.3.1 Financial Instruments . . . . . . . . . . . . . . . . . . . . . . 12
2.3.2 Sources of Return . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.3 Absolute Return Targets . . . . . . . . . . . . . . . . . . . . . 15
2.3.4 Performance Fees . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.5 Co-Investment . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.6 Liquidity of Hedge Fund Investments . . . . . . . . . . . . . . 20
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
III
Contents IV
3 Hedge Fund Strategies 25
3.1 Risks in Hedge Fund Investing . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1 Industry-Inherent Risk Factors . . . . . . . . . . . . . . . . . 26
3.1.2 Strategy-Specific Risk Factors . . . . . . . . . . . . . . . . . . 31
3.2 Hedge Fund Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Equity Hedge . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.1.1 Equity Hedge Subgroups . . . . . . . . . . . . . . . . 35
3.2.1.2 Investment Approaches . . . . . . . . . . . . . . . . 37
3.2.1.3 Equity Hedge Trades . . . . . . . . . . . . . . . . . . 38
3.2.1.4 Equity Hedge Risk Factors . . . . . . . . . . . . . . . 40
3.2.2 Convertible Arbitrage . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.3 Fixed Income Arbitrage . . . . . . . . . . . . . . . . . . . . . 49
3.2.4 Mortgage-Backed Securities Arbitrage . . . . . . . . . . . . . . 55
3.2.5 Merger Arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2.6 Distressed Securities . . . . . . . . . . . . . . . . . . . . . . . 64
3.2.7 Global Macro . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.8 Short Selling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.2.9 Emerging Markets . . . . . . . . . . . . . . . . . . . . . . . . 77
3.2.10 Managed Futures . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.2.11 Closed-end Fund Arbitrage . . . . . . . . . . . . . . . . . . . . 84
3.2.12 Depository Receipts Arbitrage . . . . . . . . . . . . . . . . . . 88
3.2.13 Regulation D . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.3 Fund of Hedge Funds . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.3.1 Advantages of Funds of Hedge Funds . . . . . . . . . . . . . . 92
3.3.2 Disadvantages of Funds of Hedge Funds . . . . . . . . . . . . 95
3.3.3 Hedge Fund Selection . . . . . . . . . . . . . . . . . . . . . . . 97
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4 Description and Analysis of Hedge Fund Data 104
Contents V
4.1 Hedge Fund Databases and Index Providers . . . . . . . . . . . . . . 104
4.1.1 Desirable Properties of Hedge Fund Indices . . . . . . . . . . . 105
4.1.2 Hedge Fund Index Providers . . . . . . . . . . . . . . . . . . . 108
4.1.3 Hedge Fund Databases and Index Methodologies . . . . . . . . 118
4.2 Biases in Hedge Fund Index Data . . . . . . . . . . . . . . . . . . . . 126
4.2.1 Biases in Hedge Fund Databases . . . . . . . . . . . . . . . . . 126
4.2.1.1 Problems with Hedge Fund Data . . . . . . . . . . . 127
4.2.1.2 Survivorship Bias . . . . . . . . . . . . . . . . . . . . 128
4.2.1.3 Selection Bias . . . . . . . . . . . . . . . . . . . . . . 131
4.2.1.4 Instant-History Bias . . . . . . . . . . . . . . . . . . 132
4.2.1.5 Stale-Price Bias . . . . . . . . . . . . . . . . . . . . . 133
4.2.1.6 Consequences of Hedge Fund Database Biases . . . . 134
4.2.2 Weighting Schemes for Hedge Fund Indices . . . . . . . . . . . 139
4.2.3 Fund of Hedge Funds Data . . . . . . . . . . . . . . . . . . . . 140
4.3 Data Selection from the Hedge Fund Index Universe . . . . . . . . . . 144
4.4 Statistical Properties of Fund of Hedge Fund Index Data . . . . . . . 148
4.4.1 Return Definitions . . . . . . . . . . . . . . . . . . . . . . . . 150
4.4.2 Unconditional Return Distributions . . . . . . . . . . . . . . . 153
4.4.2.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . 153
4.4.2.2 Parameter Estimates . . . . . . . . . . . . . . . . . . 160
4.4.2.3 Tests for Normality . . . . . . . . . . . . . . . . . . . 162
4.4.3 Time Series Analysis . . . . . . . . . . . . . . . . . . . . . . . 166
4.4.3.1 Stationarity . . . . . . . . . . . . . . . . . . . . . . . 166
4.4.3.2 Autocorrelation . . . . . . . . . . . . . . . . . . . . . 168
4.4.3.3 Unsmoothing of Hedge Fund Data . . . . . . . . . . 171
4.4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5 Portfolio-Selection including Hedge Funds 176
Contents VI
5.1 Traditional Portfolio Selection . . . . . . . . . . . . . . . . . . . . . . 177
5.2 Problems with Traditional Portfolio-Selection . . . . . . . . . . . . . 181
5.2.1 Investor Preferences . . . . . . . . . . . . . . . . . . . . . . . 181
5.2.2 Return Distributions . . . . . . . . . . . . . . . . . . . . . . . 183
5.2.3 Dependence Structures . . . . . . . . . . . . . . . . . . . . . . 184
5.2.3.1 Linear Correlation . . . . . . . . . . . . . . . . . . . 184
5.2.3.2 Empirical Evidence against Linear Correlation . . . . 187
5.2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.3 Modeling Multivariate Return Distributions . . . . . . . . . . . . . . 192
5.3.1 Marginal Distributions . . . . . . . . . . . . . . . . . . . . . . 193
5.3.1.1 Moments of Univariate Distribution Functions . . . . 194
5.3.1.2 Kernel Densities . . . . . . . . . . . . . . . . . . . . 195
5.3.2 Copulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.3.2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . 196
5.3.2.2 Important Copula Properties . . . . . . . . . . . . . 199
5.3.2.3 Different Copula Functions . . . . . . . . . . . . . . 200
5.3.2.4 Fitting Copula Functions . . . . . . . . . . . . . . . 211
5.3.2.5 Selecting Copula Functions . . . . . . . . . . . . . . 213
5.4 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
5.4.1 Technical Aspects . . . . . . . . . . . . . . . . . . . . . . . . . 216
5.4.2 Portfolio Constituents . . . . . . . . . . . . . . . . . . . . . . 222
5.4.3 Modeling Multivariate Return Distributions . . . . . . . . . . 222
5.4.3.1 Marginal Distributions . . . . . . . . . . . . . . . . . 222
5.4.3.2 Multivariate Distributions . . . . . . . . . . . . . . . 226
5.4.4 Portfolio Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.4.4.1 Simulation Approach . . . . . . . . . . . . . . . . . . 231
5.4.4.2 Minimum Risk Portfolios . . . . . . . . . . . . . . . 233
5.4.4.3 Efficient Portfolios . . . . . . . . . . . . . . . . . . . 236
Contents VII
5.4.4.4 Impact of Modeling . . . . . . . . . . . . . . . . . . 243
5.4.4.5 Benefits of Hedge Funds . . . . . . . . . . . . . . . . 250
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
6 Summary and Conclusion 257
Bibliography 262
List of Figures
1.1 Structure of this work . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Traditional investments in the investment universe . . . . . . . . . . . 9
2.2 Systematization of alternative investments . . . . . . . . . . . . . . . 11
2.3 Key characteristics of hedge funds . . . . . . . . . . . . . . . . . . . . 12
2.4 Relative return model vs. absolute return model . . . . . . . . . . . . 16
2.5 Dates and periods affection hedge fund liquidity . . . . . . . . . . . . 23
3.1 Systematization of risk factors in hedge fund investing . . . . . . . . 26
3.2 Systematization of equity strategies . . . . . . . . . . . . . . . . . . . 36
3.3 Advantages and disadvantages of funds of hedge funds . . . . . . . . 92
3.4 Typical hedge fund selection process . . . . . . . . . . . . . . . . . . 98
3.5 Important topics in a hedge fund due diligence . . . . . . . . . . . . . 102
4.1 Objectives and problems of hedge fund benchmark construction . . . 106
4.2 Problems with hedge fund data and resulting biases . . . . . . . . . . 128
4.3 Different reasons for biases and the resulting consequences . . . . . . 135
4.4 Constituents and weighting schemes of hedge fund indices . . . . . . . 141
4.5 Advantages and disadvantages of fund of hedge fund indices . . . . . 142
4.6 Performance of fund of hedge funds indices since 1997 . . . . . . . . . 148
4.7 Differences between simple and log returns . . . . . . . . . . . . . . . 151
4.8 Histograms for fund of hedge fund index returns since 1997 . . . . . . 157
VIII
List of Figures IX
4.9 Boxplots for fund of hedge fund index returns since 1997 . . . . . . . 159
4.10 Quantile-quantile plots for fund of hedge fund index returns . . . . . 163
4.11 Autocorrelation plots for fund of hedge fund index returns . . . . . . 170
5.1 Assumptions that lead to traditional portfolio selection . . . . . . . . 179
5.2 Bivariate densities with standard normal marginals . . . . . . . . . . 187
5.3 Portfolio modeling approach . . . . . . . . . . . . . . . . . . . . . . . 192
5.4 Bivariate densities of Gaussian copulas with different correlation co-
efficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.5 Bivariate densities of t-copulas with different degrees of freedom . . . 204
5.6 Bivariate densities of t-copulas with different correlation coefficients . 205
5.7 Bivariate densities of Clayton copulas with different coefficients βC . . 207
5.8 Bivariate densities of Frank copulas with different coefficients βF . . . 208
5.9 Bivariate densities of Gumbel copulas with different coefficients βG . . 209
5.10 Empirical copula function . . . . . . . . . . . . . . . . . . . . . . . . 210
5.11 Return series for different potential portfolio constituents . . . . . . . 223
5.12 Empirical distribution function (black) and corresponding cumulative
kernel densities (grey) . . . . . . . . . . . . . . . . . . . . . . . . . . 224
5.13 Marginal distributions represented by kernel densities . . . . . . . . . 225
5.14 Log-Likelihood for different degrees of freedom ν of the t-copula . . . 227
5.15 Illustration of the AD measure for a Gumbel copula . . . . . . . . . . 229
5.16 Dependence structure from a Gumbel copula . . . . . . . . . . . . . . 229
5.17 Bivariate scatter plots for stock and bond returns . . . . . . . . . . . 231
5.18 Bivariate scatter plots for stock and hedge fund returns . . . . . . . . 232
5.19 Bivariate scatter plots for bond and hedge fund returns . . . . . . . . 232
5.20 Standard deviations of all minimum risk portfolios . . . . . . . . . . . 235
5.21 Standard deviation - efficient frontier for Gaussian copula . . . . . . . 238
5.22 Standard deviation - efficient frontier for Gumbel copula . . . . . . . 238
5.23 MAD - efficient frontier for Gaussian copula . . . . . . . . . . . . . . 239
List of Figures X
5.24 MAD - efficient frontier for Gumbel copula . . . . . . . . . . . . . . . 239
5.25 Shortfall probability - efficient frontier for Gaussian copula . . . . . . 240
5.26 Shortfall probability - efficient frontier for Gumbel copula . . . . . . . 240
5.27 Minimum Regret - efficient frontier for Gaussian copula . . . . . . . . 241
5.28 Minimum Regret - efficient frontier for Gumbel copula . . . . . . . . 241
5.29 CVaR - efficient frontier for Gaussian copula . . . . . . . . . . . . . . 242
5.30 CVaR - efficient frontier for Gumbel copula . . . . . . . . . . . . . . . 242
5.31 Differences in standard deviation for the efficient frontiers (Gaussian
copula minus Gumbel copula results) . . . . . . . . . . . . . . . . . . 245
5.32 Differences in weights for the standard deviation - efficient frontiers
(Gaussian copula minus Gumbel copula results) . . . . . . . . . . . . 245
5.33 Differences in MAD for the efficient frontiers (Gaussian copula minus
Gumbel copula results) . . . . . . . . . . . . . . . . . . . . . . . . . . 246
5.34 Differences in weights for the MAD - efficient frontiers (Gaussian
copula minus Gumbel copula results) . . . . . . . . . . . . . . . . . . 246
5.35 Differences in shortfall probability for the efficient frontiers (Gaussian
copula minus Gumbel copula results) . . . . . . . . . . . . . . . . . . 247
5.36 Differences in weights for the shortfall probability - efficient frontiers
(Gaussian copula minus Gumbel copula results) . . . . . . . . . . . . 247
5.37 Differences in minimum regret for the efficient frontiers (Gaussian
copula minus Gumbel copula results) . . . . . . . . . . . . . . . . . . 248
5.38 Differences in weights for the minimum regret - efficient frontiers
(Gaussian copula minus Gumbel copula results) . . . . . . . . . . . . 248
5.39 Differences in CVaR for the efficient frontiers (Gaussian copula minus
Gumbel copula results) . . . . . . . . . . . . . . . . . . . . . . . . . . 249
5.40 Differences in weights for the CVaR - efficient frontiers (Gaussian
copula minus Gumbel copula results) . . . . . . . . . . . . . . . . . . 249
List of Figures XI
5.41 Standard deviation - efficient frontier with (black) and without (grey)
hedge funds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
5.42 MAD - efficient frontier with (black) and without (grey) hedge funds 252
5.43 Shortfall probabilities - efficient frontier with (black) and without
(grey) hedge funds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
5.44 Minimum regret - efficient frontier with (black) and without (grey)
hedge funds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
5.45 CVaR - efficient frontier with (black) and without (grey) hedge funds 254
List of Tables
2.1 Usage of leverage with different hedge fund strategies . . . . . . . . . 14
3.1 Typical equity hedge trades . . . . . . . . . . . . . . . . . . . . . . . 40
3.2 Important risk factors in equity hedge strategies . . . . . . . . . . . . 41
3.3 Typical convertible arbitrage trades and sources of profit . . . . . . . 47
3.4 Important risk factors in convertible arbitrage hedge funds . . . . . . 48
3.5 Typical fixed income arbitrage trades . . . . . . . . . . . . . . . . . . 53
3.6 Important risk factors in fixed income arbitrage hedge funds . . . . . 55
3.7 Typical mortgage-backed securities arbitrage trades . . . . . . . . . . 58
3.8 Important risk factors in mortgage-backed securities arbitrage . . . . 59
3.9 Typical merger arbitrage trades and sources of profit . . . . . . . . . 62
3.10 Important risk factors in merger arbitrage hedge funds . . . . . . . . 63
3.11 Typical distressed securities trades . . . . . . . . . . . . . . . . . . . 67
3.12 Important risk factors in distressed securities hedge funds . . . . . . . 68
3.13 Macroeconomic arbitrage strategies . . . . . . . . . . . . . . . . . . . 72
3.14 Trades involving explicit or implicit short sales . . . . . . . . . . . . . 74
3.15 Important risk factors in short selling . . . . . . . . . . . . . . . . . . 76
3.16 Emerging markets trades . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.17 Important risk factors in emerging markets hedge funds . . . . . . . . 81
3.18 Managed futures strategies . . . . . . . . . . . . . . . . . . . . . . . . 83
3.19 Important risk factors in managed futures . . . . . . . . . . . . . . . 85
XII
List of Tables XIII
3.20 Important risk factors in closed-end fund arbitrage . . . . . . . . . . 87
4.1 Summary on hedge fund indices and index providers . . . . . . . . . . 119
4.2 Summary on hedge fund index strategies I . . . . . . . . . . . . . . . 121
4.3 Summary on hedge fund index strategies II . . . . . . . . . . . . . . . 122
4.4 Summary on hedge fund database representativeness . . . . . . . . . 123
4.5 Summary on hedge fund index representativeness . . . . . . . . . . . 124
4.6 Summary on hedge fund index transparency . . . . . . . . . . . . . . 125
4.7 Survivorship bias in hedge fund databases . . . . . . . . . . . . . . . 137
4.8 Instant-history bias in hedge fund databases . . . . . . . . . . . . . . 138
4.9 Summary on reasons for hedge fund database biases . . . . . . . . . . 145
4.10 Fund of hedge fund index providers . . . . . . . . . . . . . . . . . . . 146
4.11 Fund of hedge fund index performance . . . . . . . . . . . . . . . . . 149
4.12 Descriptive statistics for fund of hedge fund index returns since in-
ception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.13 Descriptive statistics for fund of hedge fund index returns since 1997 155
4.14 Descriptive statistics for index log returns since 1997 . . . . . . . . . 156
4.15 Quantiles for fund of hedge fund index returns since 1997 . . . . . . . 158
4.16 Maxima and Minima for fund of hedge fund index returns since in-
ception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.17 Parameter estimates for fund of hedge fund index returns since 1997 . 161
4.18 P-values of different tests for normally distributed index returns . . . 164
4.19 P-values of different tests for normally distributed index log returns . 165
4.20 P-values of different tests for unit roots or stationarity of index returns168
4.21 Autocorrelation in index return time series . . . . . . . . . . . . . . . 171
4.22 Descriptive statistics for unsmoothed hedge fund index returns . . . . 173
4.23 Parameter estimates for unsmoothed fund of hedge fund index returns 173
5.1 Descriptive statistics for the estimated kernel densities . . . . . . . . 224
List of Tables XIV
5.2 Cumulative probabilities for the estimated kernel densities . . . . . . 226
5.3 Linear correlation matrix of historical returns . . . . . . . . . . . . . 226
5.4 Estimates for βC , βF , and βG . . . . . . . . . . . . . . . . . . . . . . 227
5.5 Goodness-of-fit for different copulas . . . . . . . . . . . . . . . . . . . 228
5.6 Minimum risk portfolios for different risk measures . . . . . . . . . . 233
5.7 Minimum risk portfolios without hedge funds . . . . . . . . . . . . . . 250
List of Abbreviations
AAC Alternative Asset Center
ABN/EH ABN AMRO/Eurekahedge
ABS Asset-Backed Securities
AD Anderson-Darling
ADR American Depository Receipt
APT Arbitrage Pricing Theory
ARIX Absolute Return Investable Index
AUM Assets Under Management
BaFin Bundesanstalt fur Finanzdienstleistungen
BlueX Blue Chip Hedge Fund Index
CAPM Capital Asset Pricing Model
CBAS Convertible Bond Asset Swap
CDO Collateralized Debt Obligation
CFO Collateralized Fund Obligation
CFD Contracts for Difference
CFTC Commodity Futures Trading Commission
CISDM Center for International Securities and Derivatives Markets
CMO Collateralized Mortgage Obligation
CPO Commodity Pool Operator
CSFB Credit Suisse First Boston
CTA Commodity Trading Advisor
XV
List of Abbreviations XVI
CtD Cheapest-to-Deliver
CVaR Conditional Value-at-Risk
CvM Cramer-von Mises
cw capacity weighting
DR Depository Receipt
EACM Evaluation Associates Capital Markets
ed. editor(s)
edn edition
et al. et alii
ETF Exchange Traded Funds
ew equal weighting
FHLMC Federal Home Loan Mortgage Corporation
FNMA Federal National Mortgage Association
FRM Financial Risk Management
FSA Financial Services Authority
FT Financial Times
GDP Gross Domestic Product
GDR Global Depository Receipt
GNI Gross National Income
GNMA Government National Mortgage Association
HFI Hedge Fund Intelligence
HF-NET Hedgefund.net
HFR Hedge Fund Research
IO Interest-Only
JB Jarque-Bera
KPSS Kwiatkowski, Phillips, Schmidt, Shin test
LBGC Lehman Brothers Government/Credit
Libor London Interbank Offered Rate
List of Abbreviations XVII
LKS Kolmogorov-Smirnov test with Lilliefors correction
MAD Mean-Absolute Deviation
MAR Managed Accounts Reports
MBS Mortgage-Backed Securities
MR Minimum Regret
MSCI Morgan Stanley Capital International
NAV Net Asset Value
OFD Offshore-Funds Directory
OTC Over-the-Counter
p. page
PAC Planned Amortization Class
PCA Principal Components Analysis
PIPE Private Investments in Public Equities
PO Principal-Only
pp. pages
S&P Standard and Poors
SD Standard Deviation
SE Standard Error
SEC Securities and Exchange Commission
SF Shapiro-Francia
SfP Shortfall Probability
SW Shapiro-Wilk
TED Treasury/Eurodollar-Spread
TMC The Mortgage Corporation
Ts. Taunus
UK United Kingdom
US United States
VaR Value-at-Risk
List of Abbreviations XVIII
VC Venture Capital
vw value weighting
List of Symbols
AD(.) Anderson-Darling statistic
c(.) density of the empirical copula function
C(.) copula function
C(.) empirical copula function
CC(.) Clayton copula function
CF (.) Frank copula function
CG(.) Gumbel copula function
cN(.) density of the Gaussian or Normal copula
CN(.) Gaussian or Normal copula
Cov covariance
CoV coefficient of variation
CP (.) product copula
ct(.) density of the t-copula
Ct(.) t-Copula
CVaR(.) Conditional Value-at-Risk function
fi(.) probability density function for returns of asset i
fi(.) estimated probability density function for returns of asset i
Fi(.) cumulative probability density function for returns of asset i
Fi(.) estimated cumulative probability density function for returns
of asset i
F−1(.) quantile function or inverse cumulative density function
XIX
List of Symbols XX
h bandwidth parameter of the kernel estimator
i, j indices
IAD(.) Integrated Anderson-Darling statistic
K(.) kernel function
KU sample kurtosis
KU estimated kurtosis
KUi kurtosis for returns of asset i
L lattice for the empirical copula
L(.) log-likelihood function
M(.) function for the upper probability bound
MAD(.) mean-absolute deviation function
Pt asset price at time in point t
rt continuous return for period t
rP,t continuous portfolio return for period t
ri,t continuous return of asset i for period t
Rt discrete return for period t
RP,t discrete portfolio return for period t
Ri,t discrete return of asset i for period t
RT geometric average return over period T
R∗t unsmoothed discrete return for period t
RTarget target return for the shortfall probability
RP discrete portfolio return
RminP (.) minimum portfolio return function
R vector of asset returns
RP vector of discrete portfolio returns
R vector of random discrete returns
R vector of expected discrete returns
Rs return vector for scenario s
List of Symbols XXI
S number of scenarios
SD(.) standard deviation function
SE standard error
SfP(.) shortfall probability function
SK sample skewness
SK estimated skewness
SKi skewness for returns of asset i
t index
T number of periods or number of realized historical returns
Ti(.) transformation function for random variable i
T−1ν (.) inverse of the distribution function of the univariate t-
distribution
T nρ,ν(.) cumulative density function of the n-variate t-distribution
function with correlation matrix ρ and ν degrees of freedom
ui standard-uniformly distributed random variable
U(.) utility function
V end of period wealth
V0 initial wealth
V ar variance
VaR(.) Value-at-Risk function
VaR0(.) Value-at-Risk function
w vector of portfolio weights
wi portfolio weight of asset i
W (.) function for the lower probability bound
List of Symbols XXII
α quantile corresponding to a confidence level of 1-α
βC parameter of the generator for a Clayton copula
βC parameter estimate of the generator for a Clayton copula
βF parameter of the generator for a Frank copula
βF parameter estimate of the generator for a Frank copula
βG parameter of the generator for a Gumbel copula
βG parameter estimate of the generator for a Gumbel copula
γ vector of parameters for a marginal distribution
γ vector of parameter estimates for a marginal distribution
γ (.) autocovariance function
Γ (.) gamma function
λ smoothing coefficient
µ expected portfolio return
µ sample mean
µi expected return of asset i
ν degrees of freedom of a t-distribution
ξ vector of normal inverses or inverses from a t-distribution
φ vector of copula parameters
φ vector of copula parameter estimates
ϕ(.) generator function for Archimedian copulas
ϕ[−1](.) pseudo-inverse of a generator function for Archimedian copu-
las
Φρ(.) standardized multivariate normal distribution function with
correlation matrix ρ
Φ−1(.) inverse of the standardized univariate normal distribution
function
ρ linear correlation coefficient or correlation matrix
ρ (.) autocorrelation function
List of Symbols XXIII
σ sample standard deviation
σ estimated standard deviation
σi standard deviation for returns of asset i
χ2 Chi-Square test
1 Introduction
1.1 Motivation
The hedge fund industry is moving out of the alternative corner and into the main-
stream. Investments in hedge funds have experienced considerable and steady
growth in recent years, both in terms of the number of funds and asset volume.
In the years from 2002 to 2005 net asset flows to hedge funds of 70 to 100 billion
dollars were reported on an annual basis. According to leading hedge fund data
providers, the estimated assets under management reached more than one trillion
dollars in 2005.1 This amount is invested in approximately 6100 single hedge funds
and 1800 funds of hedge funds.2
Investments from institutional investors in particular have increased considerably.
Some well-known and sophisticated investors already hold significant proportions
of hedge funds.3 While high net worth individuals are currently believed to own
two thirds of the global hedge fund capital, research institutes predict substantial
growth in the institutional area. It is expected that up to 50% of hedge fund assets
will be owned by institutional investors by 2007.4
1 Two thirds of this amount are invested in hedge funds with offshore domiciles, like the Caymans,
the Channel Islands, or the Bahamas.2 See Greenwich Associates (2004), HFR (2005), and Watson Wyatt (2005).3 For example, many universities report substantial investments in hedge funds, see Reserve Bank
of Australia (1999), p. 3.4 See Mercer Oliver Wyman (2005), p. 15.
1
1 Introduction 2
While institutional investors in some countries already allocate substantial parts of
their portfolio to hedge funds, institutional investors in Germany are rather reserved
about this new asset class.5 The German hedge fund market was stimulated by
the so-called “Investmentmodernisierungsgesetz” of 2004. Prior to this, hedge fund
managers were prohibited from operating in Germany and investors were subject to
punitive capital gains tax. In 2004 investment regulations were relaxed and single-
strategy funds and funds of hedge funds were permitted to be domiciled in Germany.
Furthermore, domestic investors were allowed to allocate money to domestic and
offshore funds. The new investment legislation gives equal tax treatment to domestic
and foreign funds, but the requirements to be met by hedge funds in order to obtain
this taxation status (so-called “tax transparency”) are relatively strict.6,7
More than 20 domestic hedge funds are currently authorized by the German “Bun-
desanstalt fur Finanzdienstleistungen” (BaFin) but by 2005 only about 1.5 billion
dollars’ worth of funds had been sold to investors in German single hedge funds and
funds of hedge funds. Institutional investors, especially insurance companies, one
important group of investors, are reluctant to invest in hedge funds as the BaFin has
placed restrictions on the amounts which may be invested in hedge funds and has
also introduced strict risk management requirements. These regulatory restrictions
apart, however, the majority of institutional investors are suspected of not being
adequately informed about how hedge funds and the strategies they employ really
work.8
5 See Greenwich Associates (2004) and Mercer Oliver Wyman (2005).6 See Ho (2005), p. 58, for a brief overview of the new German investment legislation, and
Livonius (2004) for more details on the ‘Investmentmodernisierungsgesetz” concerning hedge
funds.7 Retail products in the form of structured certificates were already on offer before the regulatory
reform. These certificates enable investors to participate in the performance of hedge fund
investments but without the tax transparency requirements. See Mader/Echter (2004) for an
overview of the German market for hedge fund certificates.8 See Ho (2005), p. 59.
1 Introduction 3
1.2 Objectives and Structure
There are several reasons for the growing interest in hedge funds. A major moti-
vation for many investors is the historically attractive return profile of such funds.
These returns have triggered massive capital inflows and, as a consequence, a need
for new funds and managers. Whether there is adequate supply to meet the growing
demand remains an open question. Market observers estimate that only approxi-
mately 5 - 10% of hedge fund managers are able to add significant value on an
after-fee basis.9 Identifying those managers who are able to provide such additional
value is probably the main task investors have to face.
In addition to this central problem of selecting skilled hedge fund managers, in-
vestors face a variety of other challenges: legal and accounting issues, regulatory
requirements, compensation structures, trading strategies, portfolio considerations,
performance measurement, etc. These issues are addressed by several research insti-
tutes10, some academic journals and also by a number of publications for institutional
investors that concentrate on hedge funds in particular and alternative investments
in general.
The challenges faced by investors form the starting point for this work, and its
main objectives are to analyze different sources of risk in hedge fund investing and
to improve the quantitative modeling of hedge fund investments in the portfolio
context. Figure 1.1 on the following page gives an overview on the different topics
covered.
9 See Watson Wyatt (2005).10 Examples of research institutes that focus on hedge funds or alternative investments in general
are the “Centre for Hedge Fund Research and Education” at London Business School, the
“Center for International Securities and Derivatives Markets” (CISDM) at the University of
Massachusetts, the “Edhec Risk and Asset Management Research Centre” at the Edhec Business
School of Nice, and the “Alternative Investment Research Centre” at the Cass Business School
of London.
1 Introduction 4
����������
���������
������������� �����
����������
����������� ����� ������������������� �
�������� �
���������������������������������������
����������
������������� ���
Figure 1.1: Structure of this work
The work covers a variety of aspects, with the four main chapters combining qualita-
tive and quantitative information on the hedge fund industry. Generally speaking,
Chapters 2, 3, and the first part of Chapter 4 provide qualitative hedge fund re-
search while quantitative aspects are discussed in the second part of Chapter 4 and
in Chapter 5.
The basic characteristics of hedge funds and alternative investments in general are
outlined in Chapter 2. In this chapter, hedge fund industry standards and the char-
acteristics typically exhibited by hedge funds are described in detail and compared
to those in the traditional investment universe. The chapter also concentrates on
the definition of hedge funds and the alternative investment universe.
After discussing the special features of the hedge fund industry, we analyze which
sources of risk may justify the historically very attractive returns achieved by the
different hedge fund strategies. Although the relevant literature contains few sys-
tematic overviews of individual hedge fund strategies and the corresponding trades,
1 Introduction 5
such an analysis of individual strategies and funds of hedge funds is essential when
investment in hedge funds is considered. In Chapter 3 systematic and unsystematic
risk factors in hedge fund investing and the risk premia captured by hedge fund
strategies are introduced. Furthermore, the relevant trades and the corresponding
risk factors for several classical hedge fund strategies are described. Since new in-
vestors in hedge funds typically choose the fund of hedge funds structure instead of
direct investment in single hedge funds11, Chapter 3 also analyzes the fund of hedge
funds structure. This way of structuring hedge fund investments offers investors
some advantages but also has some drawbacks.
Chapter 4 addresses the need for reliable hedge fund return data. When investors
consider allocating money to hedge funds they form an expectation of the develop-
ment of future returns. Since historical data is usually their starting point in this,
Chapter 4 examines in detail the most important hedge fund index data providers
and the relevant indices and takes a closer look at the different hedge fund databases
and the different index methodologies. After discussing several data biases and their
implications, the statistical properties of fund of hedge funds indices are analyzed.
Such an analysis of historical behavior is a vital issue for an investor interested in
the quantitative modeling of the risk return characteristic and the impact of hedge
funds. This in-depth analysis also forms the starting point for the analysis of a
portfolio including hedge funds.
The implications of the statistical properties of hedge fund index returns for portfolio
selection are discussed in the first sections of Chapter 5. We introduce an alternative
one-period modeling approach for portfolio returns that is flexible enough to cap-
ture these return characteristics. While Markowitz’s12 classical portfolio selection
framework involves rather strict assumptions about the structure of asset returns
11 See State Street (2005), p. 5. 35% of global hedge fund assets are currently held by funds of
hedge funds, see Mercer Oliver Wyman (2005), p. 16.12 See the original work of Markowitz (1952).
1 Introduction 6
or utility functions, the approach presented in this work is very flexible. In a first
step we model the marginal return distributions of the individual assets in order
to reproduce the univariate return characteristics of the assets under consideration.
In a second step, we focus on the dependence structure of the asset returns in the
portfolio. This modeling of dependencies with multivariate copula functions delivers
more realistic results than the classical portfolio selection framework. Combining
the marginal return distributions and the dependence structures delivers a multi-
variate model for asset returns. This model is applied in a case study which analyzes
a traditional portfolio in which hedge funds are included. This is a major issue for
hedge fund investors and, with the model for portfolio returns at hand, we are able
to derive optimal allocations to asset classes coupled with an optimization with
respect to different risk measures.
Chapter 6 concludes with a brief summary of the most important results. In addi-
tion, we make suggestions for future research.
2 Hedge Funds
The history of hedge fund investment dates back to the 1940s.1 But over the last
ten years the interest in and the depth of this market segment have increased sig-
nificantly. This has lead to a high degree of heterogeneity within the hedge fund
industry of today.2 Therefore, thorough analysis of this industry is very important.
This chapter provides an introduction to hedge funds and outlines some of the char-
acteristics of this investment opportunity. First, Section 2.1 introduces the term
“hedge fund”. In Section 2.2 a categorization of hedge funds is provided and Sec-
tion 2.3 specifies some important hedge fund characteristics that are fundamental
for the understanding of this investment opportunity.
2.1 Hedge Funds Defined
Various definitions of the term “hedge funds” circulate in the relevant literature.3
The term itself could be misleading in two ways. On the one hand “hedging”
risk is not necessarily the main focus of hedge fund managers.4 The widespread
misunderstanding that hedge funds in general pursue high risk strategies is just
1 See for example Kaiser (2004), pp. 57-73, Ineichen (2003), pp. 3-19, or Jaeger (2002), pp.
20-23, for a brief history on hedge funds.2 See Jaeger (2001), p. 3.3 See UBS (2004), pp. 112-117, for an interesting and sometimes amusing survey of hedge fund
definitions.4 See Lhabitant (2002), p. 13.
7
2 Hedge Funds 8
as false as the perception that a hedge fund manager hedges all risk factors in his
investment strategy. As (ex-ante) returns are theoretically a function of risk, hedging
all the relevant risk factors would imply an expected return on the level of riskfree
assets.5 On the other hand the declaration as “fund” could be misleading as the
legal structure is not identical to common (mutual) funds.6
As there is no exact legal definition of hedge funds there are many partly contradic-
tory definitions circulating. A very simple and sensible definition of “hedge funds”
is given in Bookstaber (2003).7 When the universe of all tradable securities and
possible trading strategies is regarded , then hedge funds are described as the whole
investment universe minus a small slice that is referred to as traditional investments.8
These traditional investment funds and/or investment strategies are characterized
by three restrictions. At first the leverage in the fund structure is limited (smaller
than 1), furthermore these funds are long only, meaning they do not employ short
selling on a big scale. Finally these traditional funds are restricted to traditional
assets like stocks and bonds that are traded in developed markets9.10
As a consequence hedge funds (or alternative investments in general) can be “non-
traditional” in two ways. Opposed to investments that consist of unlevered long
positions in stocks, bonds, money market products, or currencies there are differ-
ences regarding the asset classes invested in and/or the strategies pursued.11
On the one hand, the hedge fund manager can allocate money in alternative asset
classes such as private equity, real estate or commodities. These alternative assets
are described in the following Section 2.2. On the other hand, the classification as
5 See for example UBS (2004), p. 111.6 See Bekier (1998), p. 75.7 See Bookstaber (2003), p. 19.8 See Bookstaber (2003), p. 19.9 See Section 3.2.9 on page 77 for a detailed discussion of emerging vs. developed markets.10 See Bookstaber (2003), p. 19.11 See for example Barra RogersCasey (2001), p. 3-4, or Jaffer (2000), p. 225.
2 Hedge Funds 9
alternative can stem from the non-traditional investment strategy employed by a
fund manager. Traditional investment strategies consist of long positions in assets
and use derivatives for hedging purposes only. Reasons for this limited flexibility
in traditional investment strategies can be found in the regulatory environment for
most common funds. Alternative investment strategies offer by far more flexibility
as they combine long and short positions and make use of leverage. Within these
alternative strategies, derivatives are widely employed for hedging, for speculation
purposes and/or for providing leverage. Figure 2.1 gives a consolidated interpreta-
tion of the hedge fund definition.
������������
��������
���������
��� ��� �
��� ��� �
�� �������
�������
��� ��� �
��� ���� �����
���� ��������
�� ���� �
�����������
Figure 2.1: Traditional investments in the investment universe12
It becomes obvious that hedge funds or alternative investments in general are the
investment universe minus a small slice, which is called traditional investments. The
intersection of the restrictions in strategies and asset classes represents the tradi-
tional investment world, while the remaining investment universe is the workplace
for hedge fund managers. So what is called hedge fund is really the full spectrum
12 See the definition in Bookstaber (2003), p. 19.
2 Hedge Funds 10
of investment strategies.13 This definition also points out that hedge funds are not
a homogeneous group. We will adopt this extensive definition of the term “hedge
fund” which sets up the basis of the analysis in the following chapters.
2.2 Hedge Funds in the Alternative Investment
Universe
The alternative investment universe offers a very broad range of different assets and
products to invest in. The position of hedge funds in this part of the investment
universe shall be outlined in this section. In Figure 2.2 on the next page a system-
atization of alternative investments is given. The universe of alternative investments
can be divided in alternative asset classes14 and alternative investment strategies.
Hedge funds are members of both alternative categories.15
Single hedge funds are rather investment strategies than an asset class, as they
are very heterogeneous.16 The classification of single hedge funds and managed
futures funds17 as “alternative” usually stems from the strategies employed by a
fund manager.18 These alternative investment strategies are also commonly referred
to as “skill-based strategies”.19 The assets a fund invests in are very often exchange-
13 See also Cottier (1998), p. 17, for a resembling definition in terms of derivatives usage, short
selling and the use of leverage.14 The definition of an asset class is not that easy. At least the members/securities of such an
asset class should be homogeneous and an asset class should provide unique risk, return and
correlation characteristics, see Oberhofer (2001) and CRA Rogers Casey (2003) for a summary
of asset class properties.15 Some authors view hedge funds in general as a separate asset class, see for example Ineichen
(2003), p. 179.16 See Oberhofer (2001).17 In the following single hedge funds and managed futures funds will be summarized under the
general term “hedge fund” (see the illustration of hedge fund strategies in Section 3.2.10).18 See for example Nakakubo (2002), p. 1.19 See Jaeger (2002), p. 17.
2 Hedge Funds 11
traded bonds or stocks. But some managers also implement long-only strategies with
other alternative asset classes or less liquid privately traded products.20
��������������� ����
���������������������� ���������� ����� �������������
������������������������
����������������
����������
��� �!"� ����� �
#����� ���������� �
������"� ����� �
������
����� �����
"��$%��� &�'�
Figure 2.2: Systematization of alternative investments21
Fund of hedge funds that invest in these alternative investment strategies in order to
diversify and avoid any directional exposure to risk factors (so-called non-directional
fund of hedge funds) can be categorized as alternative asset class.22 These funds are
much more homogeneous and fulfill some important asset class features. Especially
the properties of an estimable risk profile, a low correlation with other asset classes
and an adequate liquidity are met by non-directional fund of hedge funds.23
20 Strategies with traditional long-only investments in such alternative asset classes are for example
Regulation D (see Section 3.2.13 on page 90) and Distressed Securities (see Section 3.2.6 on
page 64).21 See for example Jaeger (2002), p. 19, for a related systematization.22 See Section 3.3 on page 91 for more details on fund of hedge funds.23 See CRA Rogers Casey (2003), p. 18.
2 Hedge Funds 12
2.3 Key Characteristics of Hedge Funds
This section covers some important hedge fund characteristics that are common to
most or even all hedge funds. The partially overlapping categories that are looked at
in greater detail are the financial instruments that are employed by hedge funds, the
investment approaches, the absolute return targets of most managers, the incentive
fees that are very common and the management investment in hedge funds. The
last topic that is covered in this section is the liquidity of hedge fund investments.
See Figure 2.3 for a summary of these key characteristics.
���������������� �������������
�������������� ���� ��������
���� � �� ������� �� �� �� ������ �� ���������� ��
�� ���� ����� ���� ����
� ����� ������
� � � � �� ����� �� �������� ���� ������ ��� ���
�!��� ��� ���� � ��� ���" �����
Figure 2.3: Key characteristics of hedge funds
2.3.1 Financial Instruments
The investment approach of hedge funds is very flexible.24 In general the full range
of securities and derivatives is available to the hedge fund manager. Positions can
be set up with equity, fixed income, currency and commodity instruments. Other
24 See for example Bekier (1998), pp. 89-91.
2 Hedge Funds 13
rather new fields for hedge funds are trading or investing in credit derivatives and
structured products, like for example collateralized debt obligations (CDOs). The
financial instruments a hedge fund holds can either be traded at an exchange or
in the over-the-counter (OTC) market. Often the liquidity of hedge fund positions
is rather low in order to capture some kind of illiquidity premium25.26 The most
common strategies and the corresponding securities, trades and risk factors are
outlined in Chapter 3.
As mentioned above, most hedge funds make use of short selling and/or leverage.
While traditional money managers are usually only allowed to set up long positions,
a hedge fund is much more flexible in its investment policy. A hedge fund manager
can easily set up a negative exposure to the market, an index and/or the price of
a certain security. This is usually done by selling borrowed securities.27 The usage
of leverage differs among strategy groups and individual hedge funds. Very often
hedge funds have overall leverage limits.28 Table 2.1 gives an idea of the amount
of leverage for different hedge fund strategies and strategy groups.29 It is obvious
that these strategies employ derivatives and debt financing differently.30 Within a
hedge fund structure there are different ways to build up such a leverage. The most
obvious leverage facility is the use of external (debt-) financing to enlarge the fund’s
capital. Another way to increase a fund’s capital base are repurchase agreements,
options, swaps or futures, as these derivative contracts offer implied leverage.31
25 See Section 3.1.2 for details on liquidity risk.26 See Bekier (1998), pp. 106-107.27 Different possibilities to establish short positions or in general negative exposures are outlined
in Section 3.2.8 on page 72.28 See Cottier (1998), p. 49.29 Osterberg/Thomson (1999) report slightly lower numbers. For 85% of the hedge funds a leverage
of two or less is estimated.30 See Chapter 3 for details on these strategies and specific trades.31 See for example Kaiser (2004), pp. 30-33.32 Data: VanHedge, December 2003.
2 Hedge Funds 14
Strategies No Low Leverage High Leverage
Leverage (less than 2:1) (more than 2:1)
Distressed Securities 48.2% 45.6% 6.1%
Emerging Markets 36.3% 46.8% 16.9%
Fund of Hedge Funds 31.9% 51.0% 17.1%
Arbitrage 18.3% 22.8% 58.8%
Global Macro 11.3% 37.1% 51.6%
Total 27.0% 45.1% 27.9%
Table 2.1: Usage of leverage with different hedge fund strategies32
2.3.2 Sources of Return
With the flexibility in instruments, leverage and short selling (outlined in Section
2.3.1) at hand, hedge fund managers focus on capturing risk premia and/or on de-
tecting and exploiting market inefficiencies. Here the terminology of capital market
models is used to differentiate between the two sources of return that are usually
labeled with “beta” for the risk premia and “alpha” for returns in consequence of
inefficiencies.
A risk premium is offered in the market in exchange for taking on systematic risks
(like market risk, credit risk or liquidity risk, see Section 3.1.2 on page 31). These
risk premia are permanent price effects, only the amount of the reward per unit of
risk fluctuates over time.33 These exposures to systematic risk factors, which are
measured with the corresponding beta factors, are important sources of hedge fund
returns.34 According to Jaeger (2003) such systematic risk premia generate about
33 See Jaeger (2003).34 See for example Wallmeier (1997), pp. 56-86, or Steiner/Bruns (2002), pp. 22-37, for details
on the Capital Asset Pricing Model (CAPM) and multifactor models like the Arbitrage Pricing
Model (APT).
2 Hedge Funds 15
80 percent of the hedge fund industry’s return.
Excess returns above such a compensation for bearing risk can be attributed to
manager skill (or luck). Here the manager was able to detected opportunities where
it is possible to earn more than the relevant risk premia, for example with mispriced
securities or superior price relevant information.35 Following Jaeger (2003) the ex-
ploitation of these inefficiencies makes up for about 20 percent of the hedge funds’
returns.
2.3.3 Absolute Return Targets
The absolute return approach of (most) hedge fund managers is quite different to
the traditional relative return approach in asset management. A relative return
manager looks for opportunities to outperform passive market benchmarks or indices
in increasing and decreasing market environments. Therefore, even with negative
absolute returns a relative return strategy could be successful. This relative return
approach also fits with traditional performance evaluation techniques. Here risk is
measured as tracking or active risk. If the fund return departs from the benchmark
return this is perceived as source of risk. As Ineichen (2003) puts it the benchmarking
approach can be viewed as a method of limiting the potential for surprises, either
positive or negative.36 The main problem with this relative return approach is that
there is no incentive for the management to preserve the wealth of investors in down
markets. The relative return fund manager tolerates negative returns brought about
by the market volatility and leaves risk management (besides the tracking risk) to
the investor.37 The main differences between classical relative return and absolute
return funds are outlined in Figure 2.4 on the following page.
35 See Jaeger (2002), pp. 24-25.36 See Ineichen (2003), p. 19.37 See Ineichen (2004), p. 23.38 See for example Ineichen (2004), p. 23.
2 Hedge Funds 16
��������
������ ���
����������
����� ����������������
��� ���������
� ���
� �� �� �� ���� �
� ����� ������
������������
������� ���
��� ��������
�����
��� ��� ���� �
����
���������
!��������� ��
�� ��� ���� �
�����
�������
��������������
��
��
Figure 2.4: Relative return model vs. absolute return model38
The absolute return approach tackles some of the problematic issues with the relative
benchmark perspective.39 The objective of absolute return managers is to earn a
consistent rate of return regardless of the bond or stock market environment.40
Typically hedge funds have such absolute return targets.41 Here the interest of
investors for capital preservation and steady profits without relying on the market
environment is aligned with management interest by incentives for the hedge fund
managers that are outlined in sections 2.3.4 and 2.3.5. Therefore, very common
return benchmarks for hedge funds are the returns (or multiples) of risk free assets
like the three-month T-bill sometimes with a risk premium on top.42 These returns
are often reflected in a fund’s hurdle rate.43 This focus on absolute, positive returns
39 A brief history of the absolute return approach can be found in Ineichen (2003) on pages 2-19.40 See for example Lhabitant (2002), p. 18, Stonham (1999), p. 285, or Schneeweis et al. (2002),
p. 7.41 See Fung/Hsieh (1999b), p. 321.42 See for example Ineichen (2003), p. 19, and Spitz (2001), p. 60.43 For more details on hurdle rates, see Section 2.3.4, p. 19.
2 Hedge Funds 17
entails a different understanding of risk. Contrary to the relative return world the
relevant risk for an absolute return manager is total risk, which describes the problem
of facing uncertain absolute results.44
2.3.4 Performance Fees
A characteristic feature of hedge funds is the incentive system for the management.
In order to attract the industry’s best managers there are very interesting com-
pensation schemes in place for hedge fund managers.45 While the compensation
of traditional fund managers is solely based on a management fee as percentage of
assets under management, hedge funds usually charge additional performance or
incentive fees.46 As the management fees for hedge funds are essentially intended to
meet operating costs, the incentive fees are meant to encourage managers to achieve
maximum returns or at least to perform well.47 The size constraint48 in some hedge
fund strategies gives another reasoning for performance fees. As the management
should not increase asset size by any means in order to collect management fees there
must be an incentive to produce positive returns and resist net growth.49 According
to Grinold/Rudd (1987) these performance or incentive fees in general are suitable
for unskillful investors and good or at least average managers. But bad managers
and sophisticated investors should rather avoid such performance fees.50
The option-like payoff structure within this compensation scheme might be prob-
lematic, as the manager participates in profits but not in losses. If the fund performs
44 See Ineichen (2004), p. 23.45 See Tarrant (1996), p. 146.46 Under various jurisdictions incentive fees are not allowed for traditional investment funds, see
Cottier (1998), p. 36. But they are prevalent in the real estate sector and the venture capital
sector.47 See Lhabitant (2002), p. 17.48 See Section 2.3.6.49 See for example Goetzmann et al. (2003), p. 1716.50 See Grinold/Rudd (1987), p. 37.
2 Hedge Funds 18
well, the manager is able to collect the additional performance fee. Conversely, if
the fund performs poorly, the manager collects the management fee.51 This sets an
incentive for the manager to increase the volatility of the fund’s net asset value in
order to maximize fees. Especially when the option-like incentive contract is out of
the money, managers have a strong incentive to increase the fund’s volatility.52 Here
the co-investment outlined in Section 2.3.5, a high watermark and hurdle rates are
able to reduce these negative side effects to some extent.53
Usually a hedge fund with a performance fee has also a so-called high watermark.
With this clause the manager commits to charging performance fees only if past
losses have been recovered.54 Therefore, the performance fee is only charged on
value added to the investor.55 These high watermark provisions in hedge fund con-
tracts limit the value of the optionality of performance fees for the management,
as the exercise price is determined by the watermark. As this combination of per-
formance fees and high watermarks has the effect of penalizing losses this has a
moderating effect on risk-taking.56 But it usually can not discourage excessive risk
taking when the fund is already substantially below the historic watermark.57 In
this case investors usually withdraw capital from the fund. This is consistent with
the common (voluntary) termination of funds when there is no reasonable possibil-
ity of meeting the high watermark provision in the incentive contract.58 Based on
51 See Bailey (1990), p. 36.52 See Brown et al. (2001), p. 1870.53 For a detailed analysis of these moral hazard problems and related issues, see Signer (2003),
pp. 119-194.54 See Cottier (1998), p. 37, for an example on the calculation of performance fees with new
subscriptions and redemptions influencing a fund’s net asset value.55 See Ineichen (2000), p. 86.56 In Goetzmann et al. (2003) the authors provide a closed-form solution for the high-water
mark contract under certain model conditions. The results show that managers have a smaller
incentive to take risks than without a high watermark.57 See Fung/Hsieh (1997b), p. 38.58 See Brown et al. (2001), p. 1882.
2 Hedge Funds 19
research by Liang (1999), such a high watermark is a determinant of fund perfor-
mance as hedge funds with high watermarks show significantly higher returns than
funds without.59
The hurdle rate of a hedge fund is another mechanism to align the hedge fund
manager’s interest with those of the investors.60 Such a hurdle rate is a rate of
return that has to be achieved before the performance fee is paid. Only above this
hurdle rate a hedge fund manager begins taking incentive fees. Here a soft or hard
hurdle can be installed in the incentive contract. A soft hurdle rate allows incentive
fees on the gross fund performance, while with hard hurdle rates the incentive fee
is calculated on the fund performance above the hurdle. As an adequate minimum
return often a reference rate, like the three-month T-bill, or a stock market index
return, like the S&P 500, is employed. But also fixed rates are common as hurdle
rates.61 This hurdle rate should ensure some kind of minimum return for the investor.
However historically most hedge funds did not have such a hurdle rate.62 The
study of Liang (1999) did not find significant influence of hurdle rates on fund
performance.63
2.3.5 Co-Investment
On the one hand the hedge fund management should act in the interest of the
investors, but on the other hand the managers try to maximize their own wealth.
With the performance fee introduced in Section 2.3.4 the investor has to account for
the problem of moral hazard.64 As the management has an incentive to maximize
the fee revenue, it is tempted by an option-like compensation with limited downside
59 See Liang (1999), p. 75.60 See Ineichen (2000), p. 87.61 See Cottier (1998), p. 36.62 See the descriptive statistics in Liang (1999), p. 75, or Ineichen (2000), p. 13.63 See Liang (1999), p. 75.64 See Signer (2003), pp. 128-132.
2 Hedge Funds 20
risk like the performance fee. A reasonable strategy for this compensation scheme
could be either alternating positive and negative returns in order to capture the
performance fee for up-movements or excessive risk-taking as outlined in Section
2.3.4. And after a period of good performance the manager might be tempted to
avoid any risk-taking until the incentive fee is paid.65
In order to align the interest of management and investors and to avoid the outlined
problem of moral hazard, managers usually make significant investments in the
fund they operate. According to data from VanHedge for the year 2003 about 82%
of the managers have a minimum of $500,000 invested in their fund.66 With an
investment in their own fund, managers are interested in protecting the fund from
(large) downside movements as their own capital is at stake as well.67 Especially in
cases where the track record of the fund is very short this is a confidence-building
measure that is able to align economic interests between the manager and external
investors.68
2.3.6 Liquidity of Hedge Fund Investments
The life-cycle of hedge fund investing is another characteristic feature of this indus-
try. From the initial investment to the final disinvestment there are some character-
istics of hedge funds that have to be considered by investors as these features affect
liquidity. In this section we will introduce minimum investments, subscription and
redemption frequencies, and capacity constraints.
A minimum investment limit is common with hedge funds.69 A (historical) rea-
65 See Lhabitant (2002), p. 18.66 This number is stable over time. In 1995 78% of the managers had a minimum investment of
$500,000, see Van (1996), p. 223.67 See Bekier (1998), p. 94.68 See Fung/Hsieh (1999b), p. 318, and Bekier (1998), p. 95.69 See for example Financial Services Authority (2002a), p. 8.
2 Hedge Funds 21
son for this fact can be found in the US regulation of hedge funds. Depending on
the structure of the hedge fund, the number of US investors is limited. Therefore,
only a minimum investment is able to ensure a sizeable pool of money even with
a smaller amount of customers.70 Under some jurisdictions, especially in Europe,
explicit minimum investment limits for hedge funds have been determined by regu-
latory authorities.71 According to data from the index provider HFR 35 percent of
all hedge funds require minimum investments above $500,000.
Subscriptions in hedge funds are restricted to certain dates. In the case of closed-
end funds investors can deposit money only during the initial issuing period or if
the fund is reopened later on for additional investments. The subscription possibil-
ities with open-end funds are much more frequent. Open-end funds usually allow
investors to deposit money at fixed dates, mostly at the end of each month. These
subscriptions are usually based on the latest net asset value.72
Similar to the subscription the redemption of hedge fund investments is also limited
to certain dates.73 Most funds allow a monthly redemption or at least multiple
redemption opportunities each year.74 After depositing money, investors of most
funds have to wait a certain period of time, the so-called lock-up period, before
the investment can be withdrawn again. On average this lock-up period is around
80-90 days.75 But some hedge funds do allow the withdrawal of money within the
lock-up period against a lock-up provision.76 When the lock-up period is over, many
70 See for example Bekier (1998), p. 95, and Cottier (1998), pp. 37-38.71 Such minimum investments have for example been installed in the investment laws of France,
Ireland or Italy. For a survey of minimum investment amounts see PriceWaterhouseCoopers
(2003), p. 13.72 See Cottier (1998), p. 41.73 Some hedge funds have a so-called key-man clause, giving the right to terminate the agreement
immediately if an important manager is leaving the fund, see Kaiser (2004), p. 99.74 See for example Ackermann et al. (1999), p. 840.75 See for example Liang (1999), p. 74, and Ineichen (2000), p. 13.76 See CRA Rogers Casey (2003), p. 6.
2 Hedge Funds 22
funds require a written advance notice to the administrator prior to a redemption.
The corresponding notice period extends the whole redemption period in order to
reduce managing cost and ensure the necessary liquidity in the fund.77 The length of
the required notice period is on average around 30 days.78 Finally the redemption
period until the actual redemption of the invested money completes the so-called
restriction period79 that also includes the notice period. This restriction period
can be interpreted as a measure for the liquidity of the hedge fund investment.80
Figure 2.5 on the next page illustrates the different dates and periods from the initial
hedge fund investment until the final disinvestment.
As mentioned before, the restrictions for redeeming hedge fund investments shall
reduce managing cost and ensure a long-term focus of investors. Scarce possibili-
ties for redemption result in lower cash reserves and should therefore increase the
performance of the fund.81 Empirical evidence on this hypothesis can be found for
example in Cottier (1998) and Das (2003).82 As hedge funds very often operate in
illiquid market environments, the nature of hedge fund investments implies some
degree of illiquidity, too.83
The recent studies by Agarwal et al. (2004) and Getmansky (2004) point out that
the performance of hedge funds is dependent on asset size and that an optimal
asset size of hedge funds does exist, especially when they hold illiquid assets, have
limited market opportunities and a high market impact of trades.84 Examples are
especially found in the strategy-groups convertible arbitrage or emerging markets.85
77 See Amin/Kat (2002b), p. 9.78 See for example Liang (1999), p. 74, and Ineichen (2000), p. 13.79 See Agarwal et al. (2004), p. 3.80 See for example Das (2003), p. 19 and p. 38, and Ineichen (2000), p. 91.81 See Lhabitant (2002), pp. 16-17.82 See Cottier (1998), pp. 188-189, and Das (2003), pp. 42-46.83 See Ineichen (2000), p. 91, and Section 3.1.2 on page 32.84 See Agarwal et al. (2004), p. 24-29, and Getmansky (2004), pp. 25-32.85 See Getmansky (2004), p. 29.
2 Hedge Funds 23
����������������
��������������
������������
����� ��������
����������� ����
�������������
������������
����� ����������
��������� ��
������ ��
Figure 2.5: Dates and periods affection hedge fund liquidity
These findings are consistent with the very common capacity constraints in the
hedge fund industry. As the opportunities for hedge fund managers in terms of
mispricings or very high risk premia are limited, most hedge funds that pursue
special trading strategies close their fund to new investors when they reach a certain
size. These limits to the investment strategies ensure an adequate return for the
investors.86 In this respect hedge fund managers behave differently from traditional
money managers. In the case of common mutual funds positive performance is
usually followed by dramatic capital inflows. With hedge funds this pattern can not
be revealed.87
86 For a related discussion of marginal returns in the hedge fund industry see Deutsche Bundesbank
(1999), pp. 34-37.87 See for example Goetzmann et al. (2003), pp. 1711-1714.
2 Hedge Funds 24
2.4 Summary
The key characteristics of the hedge fund industry or alternative investments in
general introduced in this chapter are only the first step into an interesting part
of the investment universe. These distinctive features are common to most hedge
funds, but the following Chapter 3 will point out that the hedge fund universe is
not that homogeneous. There exists a broad variety of different trading approaches
within this hedge fund universe, and the resulting risk (and return) characteristics
are manifold. Therefore, thorough analysis of hedge funds has to focus on the
different traded assets and the corresponding trading strategies in order to give a
complete picture on this part of the investment industry. Such a detailed analysis
of the most important hedge fund strategies is part of the following Chapter 3.
3 Hedge Fund Strategies
Often investors consider hedge funds as a homogeneous investment class. But as
already mentioned in Chapter 2, not all hedge funds are of similar type. The hedge
fund universe consists of funds with completely different trading approaches. These
manifold active investment strategies of hedge funds show very different risk profiles.
In this chapter the different risk factors in hedge fund investing will be introduced
and the implications for various strategies are documented. At first we will describe
some central risk factors which originate from the hedge fund industry characteristics
itself. Then common risk factors in hedge fund trading strategies will be specified as
these factors determine the profile of alternative investment strategies. In Section
3.2 some important hedge fund strategies are introduced and the sources of risk and
return are addressed. The last section of this chapter deals with fund of hedge funds.
These funds invest in target hedge funds. Section 3.3 discusses some important
characteristics and outlines the investment process of a fund of hedge funds.
3.1 Risks in Hedge Fund Investing
There are two major sources of financial risk in hedge fund investing. On the one
hand risk arises from the hedge fund sector and its special characteristics.1 In the
following these sources of uncertainty are called industry-inherent risk factors.
1 See Chapter 2 for some important characteristics.
25
3 Hedge Fund Strategies 26
On the other hand some risk factors in hedge fund investing are associated with
hedge fund trading strategies and the resulting exposures. These sources of risk are
aggregated as strategy-specific risk factors. The most important risk factors
that are introduced in the following sections are systematized in Figure 3.1.
�������������� ������������������
��������������� � �� ��� �� � �� ����
� � ��� ���� ��� ���
� �� ��� ��� ���
� � ���� ���
� ��� �� ��� ���
� � � �� ����� ���
� �� ��� ���
� � �� ��� � ��� ���
� � ����� ���
� � ����� ���
�! �������� ���
� �� ���� ���
" ������� ���
Figure 3.1: Systematization of risk factors in hedge fund investing
3.1.1 Industry-Inherent Risk Factors
Before we introduce special investment strategies, the risk factors of the alternative
investment industry in general have to be considered. This industry-inherent risk is
to a large extent due to the information asymmetry between investors and managers.
Therefore, the limited transparency in the hedge fund industry is one of the central
sources of risk.2
The industry-inherent risk factors in hedge fund investing are usually independent
of the strategies applied by the fund management. In a sufficiently broad hedge
2 See Jaeger (2002), p. 138.
3 Hedge Fund Strategies 27
fund portfolio (like a fund of hedge funds3) these risk factors can be diversified to
some extent. But in a smaller portfolio with capital allocation in a few hedge funds
only, these sources of risk have to receive a great deal of the investor’s attention and
must be controlled and managed actively.4 In the following we will introduce some
of the most important risk factors in the alternative investment industry, but this
list is not exhaustive.5
As mentioned before, lack of transparency is perhaps the central source of risk
in hedge fund investing. It is the source for many of the risk factors in hedge fund
investing. While only a permanent and full disclosure of hedge fund positions would
prevent investors from risk factors like strategy or fraud risk, information especially
on current positions is held back by the hedge fund managers.6 A possible solution
to this transparency problem is that the hedge fund discloses the current exposure
to different risk factors or at least that the fund manager reveals his (historical)
positions with an appropriate time lag.7 Managed accounts, where the investor has
turned over direction of his account to a manager, can also reduce this intrans-
parency. This set-up provides the investor with full disclosure of the trades and
therefore it addresses the basic problem of transparency and the resulting problems
of strategy and fraud risk.8 But too much transparency might also pose a threat to
a hedge fund. When confidential information about the fund’s significant positions
reaches other market participants this entails market risk for the fund.9 Therefore,
transparency that reveals the hedge fund’s risk profile regarding some predefined
risk factors but no information about individual positions or trades is a suitable
3 See Section 3.3 on page 91 for details on fund of hedge funds structures.4 See Patel et al. (2002), p. 89.5 Further potential risk factors that can be subsumed in the following categories, are mentioned
for example in Jaeger (2002).6 See Jaeger (2002), p. 138.7 See Ineichen (2000), p. 91.8 See Jaeger (2002), pp. 222-223.9 See Reynolds Parker/Warsager (2000), p. XXVI.
3 Hedge Fund Strategies 28
instrument for risk reduction.
When constructing a portfolio with allocations to hedge funds, the investor expects
a certain risk and correlation or dependence profile concerning other investments.
A change in the hedge fund’s strategy focus will expose the investor to a new set of
risk factors and is therefore a general source of risk, the so-called strategy risk.10
A typical example for such a style drift is the equity market neutral fund gaining
directional exposure or moving to new market sectors.11 A further source of strategy
risk is the leverage which the manager employs to pursue his investment strategy.12
The probability for changes in the strategy or leverage is especially high in times of
bad performance or changing markets and after the drop out of key personell.13
The lack of transparency in the hedge fund industry also leads to so-called fraud
risk. The unregulated industry facilitates fraud in many ways like for example
illegal pyramid schemes or false performance reports.14 According to SEC numbers,
there was a maximum of 5 cases of fraud per year from 1998 through 2001 and
12 documented cases in 2002, so fraud is a rather rare event.15 The detection of
fraud by the hedge fund management is a very challenging task faced by hedge
fund investors. Third party verification and auditing have to ensure the accuracy of
reported numbers and facts.16
Management risk or human risk is especially important for hedge funds as they
are very often lead by small teams or even individuals.17 So the return and risk
profile can be seen as a function of the managers’ skill.18 There are three main
10 See Moix (2004), pp. 10-11.11 See Patterson (2003), p. 46.12 See Moix (2004), p. 11.13 See Cottier (1998), p. 47.14 See Jaeger (2002), p. 140.15 See Rettberg (2003), p. 37.16 See Patterson (2003), p. 46.17 See Patel et al. (2002), p. 93.18 See Jaeger (2002), p. 138.
3 Hedge Fund Strategies 29
aspects of human risk. The first and most important source of management risk
is the ability of the management to pursue a really successful trading strategy.19
The length of a management’s (successful) track record might serve as proxy for
measuring this ability. Secondly an adequate compliance system and an attractive
incentive structure should be in place in order to motivate the management and
to avoid irrationalities. Key-men (or even key-man) dependency is a third aspect
of management risk. One or more head traders leaving the hedge fund could be
unfavorable to the fund’s perspective when the fund’s expertise in pursuing a special
strategy shrinks to zero. This problem could be encountered by long term incentive
fees or the implementation of a proprietary trading software.20
Fund age is also influences the risk profile of a hedge fund investment. Survival
risk addresses the problem that young funds (2-3 years of age) have a significantly
higher attrition rate than established funds.21 New entrants to the hedge fund
industry usually face a strong incentive to implement very risky strategies with their
performance-based funds. An option-like payoff to the manager rewards successful
high-risk bets while the manager’s downside only consists of closing the fund and
starting a new venture.22 As setting up and funding an investment platform needs
special organizational skills and the applied trading strategy (or strategies) has to
be successfully implemented in a new environment, there is also a high level of risk
for young funds not to survive the first years of their independence due to structural
or organizational problems.23
The larger the equity base of a hedge fund is getting, the more difficult it usu-
ally becomes to earn an acceptable or promised return on the employed capital.24
19 See for example Rutkis (2002), p. 46.20 See Patel et al. (2002), p. 93, and Cottier (1998), pp. 46-47.21 See for example the study of Howell (2001).22 See Section 2.3.4 on page 17 for more details on performance fees.23 See Patel et al. (2002), p. 94.24 See Tarrant (1996), p. 147.
3 Hedge Fund Strategies 30
This problem, so-called size risk, arises especially when arbitrage, convergence or
relative-value strategies are implemented.25 If the manager invests in new positions
in order to accommodate the capital inflow, the fund’s return may decline as subop-
timal investment decisions are made. When adding to existing positions, the capital
becomes less liquid as negative size effects are associated with the market impact
of large transactions.26 In order to downsize and to remain flexible, some man-
agers avoid such problems by returning capital to their investors.27 As managers
will tend to close their program when certain strategy immanent capacity limits are
reached, some hedge fund investors might also face the problem of not being able to
participate in high quality hedge funds.28
Operational risk comprises a diverse set of risks and can be broadly defined as risk
of losses resulting from inadequate or failed internal processes, people and systems,
or from external events.29 One aspect of the human factor in operational risk was
already mentioned above where management risk was discussed. Further influences
on operational risk are back-office problems, money transfer risk, clearance risk and
systems risk.30 As hedge funds very often do have a higher transaction turnover and
fewer back-up staff, operational problems, like for example limit exceedance or erro-
neous executions, are more likely than with traditional investment funds.31 Because
of the sophistication of the strategies and their timely nature, technical systems and
computing infrastructure are also extremely important for hedge funds.32 Especially
in the offshore domiciles of hedge funds basic infrastructure requirements are not
25 For directional strategies the opposite may hold due to the market impact of large traders.26 See Tarrant (1996), p. 147.27 See Cottier (1998), p. 48.28 See Jaeger (2002), p. 140, and Section 2.3.6 on page 20 on page 23.29 See Basel Committee of Banking Supervision (2003), p. 2.30 See Cruz/Davies (2000) for some examples of operational risk events, Cruz/Davies (2000), p.
134.31 See Jaeger (2002), p. 138.32 See Patel et al. (2002), p. 93, and Bousbib (2000).
3 Hedge Fund Strategies 31
always met.33
3.1.2 Strategy-Specific Risk Factors
When looking at the resulting net positions from manifold hedge fund trading strate-
gies several well known risk factors can be identified. This broad range of risk factors
is the source for risk premia which a hedge fund manager looks for. According to
Jaeger (2003) these risk premia account for 80 percent of hedge fund returns.34 Tak-
ing these systematic risks is rewarded with higher (expected) returns but the ex post
realization of the risk premia is uncertain.35
Besides traditional risk factors, like market or credit risk, there are various sources of
risk that are typical for or very pronounced in hedge fund investment strategies, for
example event or model risk. On the following pages we will introduce the central
risk factors in most hedge fund strategies: market risk, liquidity risk, credit risk,
event risk and model risk.36
Arising from unexpected (adverse) changes in the market prices or rates of financial
instruments, market risk is the most common risk factor in the investment indus-
try.37 Potential movements in the prices of equity, interest rates, credit spreads,
currency and commodities markets are captured by this risk category.38 Regarding
market risk, we can further differentiate between specific and systematic sources of
risk. While specific risk focuses on price changes of individual securities, the system-
atic component refers to general changes in market prices. As there is the possibility
33 See Jaeger (2002), p. 142.34 The remaining 20 percent earned by fund managers are contributed by the detection and ex-
ploitation of market inefficiencies, see Jaeger (2003).35 See Ilmanen et al. (2004) for the size of different risk premia over the period 1985-2002.36 As it is not possible to define all the risk categories without intersection there will be partially
overlapping risk factors in this listing.37 See for example Jorion (1997), p. 14.38 See Caxton Corporation et al. (2000), pp. 14-17.
3 Hedge Fund Strategies 32
of net short positions in financial instruments, a negative exposure to various market
prices may be the case with hedge funds (as opposed to traditional long-only mutual
funds). Changes in dependency structures, for example correlations, or volatility of
returns or market prices are also included in this category.39 Regarding hedge funds
in comparison to traditional investment funds the impact of market risk is much
more severe as hedge fund strategies very often make intense use of leverage.40
Liquidity risk is one (or the) central risk factor with hedge fund investing.41,42
Liquidity risk could be categorized as a sub-component of general market risk.43
But as it is a central risk factor to most hedge fund strategies, we will treat it as
a separate risk category. Liquidity risk can be split into funding or cash flow risk
and market liquidity risk.44 Funding liquidity risk refers to the possibility that a
(financial) firm will be unable to meet its obligations as they come due because
of an inability to liquidate assets or obtain adequate funding.45 The hedge fund
investors are a source of risk as they are also influencing the fund’s liquidity. Short
term investors with a significant stake in the fund could cause harm to long term
strategies when their capital is recalled. Therefore, a broad diversification of the
client base and/or sufficient lock-up periods are very important for hedge funds.46
Market liquidity risk arises when financial managers become unable to adjust a
position in a timely manner at a reasonable price (for example as bid-ask spreads
widen).47 As less liquid positions offer premiums, they are frequently a (major) part
39 See Duffie/Singleton (2003), p. 4.40 See Cottier (1998), p. 44, and Section 2.3.1 on page 12.41 See Rahl (2000), p. 15.42 See for example Blanco et al. (2005) on the important issue of liquidity risk modeling.43 See for example Caxton Corporation et al. (2000), p. 13, or Estenne (2000), p. 41.44 See for example Jorion (1997), pp. 15-16.45 Risks from margin calls or from the redemption policy of hedge funds belong to this risk-class
as well. See for example Cottier (1998), p. 46.46 See Chait (2000), p. 7.47 See Jameson (2001), p. 2. This part of liquidity risk could also be categorized in the market
risk subgroup. See for example Caxton Corporation et al. (2000), pp. 14-17.
3 Hedge Fund Strategies 33
of hedge fund portfolios and the resulting risk profile.
Credit risk originates from the possibility that a borrower (issuer risk) or counter-
party will fail to perform on an obligation. It can be defined as risk of default or
reduction in value associated with unexpected changes in credit quality of issuers
or counterparties.48 Counterparties include debt and equity holders. The risk from
investors as counterparties withdrawing funds is categorized among liquidity fund-
ing risk. As funding support and collateral management are vital issues for hedge
funds, changes in policies of prime brokers could have devastating consequences to a
fund.49 Here the fund manager has to ensure that the counterparty to a transaction
will be able to fulfil its end of the obligation.50 The hedge fund manager should be
aware of, mitigate and diversify this risk component.51 Sovereign risk as well as the
risk from potential changes in credit quality of clearing brokers and the probabil-
ity of default of counterparties in OTC trades are also included in this category.52
Deduced risk factors like the widening of credit spreads are subsumed in this risk
class.
As most of the hedge fund strategies are set up in very special market segments,
they are sensitive to events in their market environment. Examples for the so-
called event risk are prepayments with mortgage-backed securities arbitrage (see
Section 3.2.4 on page 55), shareholder denials or management decisions in the case
of merger arbitrage (see Section 3.2.5 on page 60), but also extraordinary political
or economic events.53 Legal risk is also summarized under event risk. Legal risk
considers different influences like court orders, changes in tax laws, interventions
from regulatory authorities and prospectus risk. Such decisions and conditions are at
48 See Duffie/Singleton (2003), pp. 3-4.49 See the example in Jaeger (2002), p. 139.50 See Press/Stamatelos (1996), p. 207.51 See Chait (2000), p. 7.52 See for example Jorion (1997), p. 15, or Cottier (1998), p. 45.53 See Jaeger (2002), p. 137.
3 Hedge Fund Strategies 34
the bottom of many hedge fund strategies54 and these events are therefore primarily
responsible for the success of such transactions. The extent of this event risk depends
on the traded securities and implemented strategies. In most cases it is simply
impossible to quantify this risk factor with mathematical models since it is based
on idiosyncratic events.55
Many arbitrage trades are identified and/or hedged with quantitative models. The
crucial point is that these models must accurately predict pricing relationships. The
model risk originates from imperfect or unrealistic assumptions, when a quantita-
tive model does not build on realistic estimates or when it is not properly reassessed
and adjusted to market conditions. As a definition it is the risk that theoretical
models used in pricing, hedging or estimating risk will produce misleading results.56
According to Allen (2003) the reasons for this risk factor can be found in the incor-
rect implementation of models, in ignoring key sources of risk during the modeling
processes or in using uncertain input parameters to calculate the model prices.57 As
many hedge fund trades are identified using models, this is an important factor in
the fund’s risk profile.
3.2 Hedge Fund Strategies
This section introduces classical hedge fund strategies. The strategies that are ex-
amined in the following paragraphs cannot give a complete summary of all possible
hedge fund trades. Rather there will be given a review for the most important hedge
fund strategies and the resulting positions.58 After a brief strategy description we
54 See for example distressed securities (Section 3.2.6 on page 64) or regulation D arbitrage (Sec-
tion 3.2.13 on page 90).55 See Bookstaber (2003), p. 21.56 See Allen (2003), p. 97.57 See Allen (2003), pp. 102-110.
3 Hedge Fund Strategies 35
will outline major trades and describe the relevant risk factors/risk premia for the
respective hedge fund category.
3.2.1 Equity Hedge
3.2.1.1 Equity Hedge Subgroups
Equity hedge is a very broad hedge fund category. According to numbers from
the index provided HFR for the first quarter of 2005 approximately one third of
the assets under management in hedge fund strategies is invested in equity hedge
strategies. These strategies in general combine long holdings of equities with short
positions in stocks.59 The equity hedge universe can be divided in a number of
subcategories. Based on the net position of the resulting equity hedge portfolio, the
different hedge fund subcategories are equity market neutral, long/short equity and
long or short biased strategies, see Figure 3.2 on the following page. The analysis of
the equity non-hedge strategy short selling can be found in Section 3.2.8 on page 72.
Levered long-only funds are very similar to traditional long-only mutual funds. In
addition to long positions in stocks these funds employ leverage to increase their
capital base.
In equity hedge approaches managers purchase stocks that are considered relatively
cheap, while simultaneously selling short stocks that are considered expensive. The
main difference between these strategy subgroups is the degree to which the resulting
portfolios are hedged concerning overall equity market risk.60
Equity market neutral strategies are designed to profit from equity market in-
58 Niche strategies like trading in energy-, weather- and credit-derivatives, options arbitrage, vot-
ing versus non-voting shares, or stub trading will not be covered in this section.59 Often stock (index) derivatives are also used to set up the necessary exposures, see Ineichen
(2001), p. 9, and Mogford (2005), pp. 31-32.60 See Parnell (2001b), p. 62.
3 Hedge Fund Strategies 36
��������������
������������ ����� �������� � � ������ �����
�������� � � ��� �� ��
�������� � � �� �� � �
� � � ���� � ��� ���
�� � ������� �
� � �� � � ��� � ��
Figure 3.2: Systematization of equity strategies
efficiencies with zero or negligible exposure to the market. The market neutrality
(beta-neutrality, currency-neutrality or both) is achieved by balanced long and short
equity positions in the hedge fund’s portfolio.61 In contrast to other equity hedge
strategies, an equity market neutral investment approach requires frequent rebal-
ancing in order to sustain the desired neutrality.62
Long/short equity hedge fund managers maintain net market exposures. They
have a more flexible investment policy as they adjust their net exposure (long or
short bias) to the equity market risk depending on the manager’s preference and the
prevailing or expected market conditions.63 Depending on the market conditions
their portfolios may be anywhere from net long to net short. A net long position
in rising markets and a net short portfolio in bear markets would be the optimal
use of this additional flexibility.64 So-called sector specialists are a special type of
61 See for example Ineichen (2000), p. 31.62 See Lhabitant (2002), p. 82.63 For a discussion of advantages and drawbacks of long/short compared with long-only strate-
gies see Michaud (1993), Arnott/Leinweber (1994), Michaud (1994), Jacobs/Levy (1995),
Jacobs/Levy (1996), Jacobs/Levy (1997), Brush (1997), Jacobs et al. (1998), Ineichen (2001)
and Kao (2002).64 See Ineichen (2001), p. 9.
3 Hedge Fund Strategies 37
long/short equity funds. Here the manager analyzes and exploiting opportunities in
some industry sectors only.65
Investment strategies with fixed long or short bias have a dedicated exposure to
one side of the equity market and are therefore less flexible than general long/short
equity approaches.
3.2.1.2 Investment Approaches
There are three basic approaches to portfolio composition in equity hedge investing:
fundamental analysis, quantitative multifactor analysis and statistical arbitrage.
Fundamental analysis is the tool for investors with a rather long-term invest-
ment horizon. With fundamental analysis the stock selection is discretionary and
aims at the identification of fundamental mis-valuations.66 Quantitative but also
qualitative factors, like for example accounting practices and management ability or
credibility have important influence on the investment decision.
A second approach to equity hedge investing is the use of factor models to iden-
tify or trade risk premia. Quantitative multifactor analysis looks at risk or return
factors that affect or explain stock prices. The detection of under- and overvalued
securities with quantitative research is often based on value or growth investing
strategies.67 Other investment approaches try to exploit well known anomalies like
the size effect68, the book-to-market effect69 or human irrationalities known from
behavioral finance70. These investment styles or return factors can be found in most
implementations of stock picking algorithms for detecting over- and/or undervalued
65 See Ineichen (2003), pp. 305-319, and Lhabitant (2002), pp. 118-119.66 See Blatter (2003), p. 14.67 See for example Reiss (1996) or Nicholas (2000), pp. 223-225, for further details on these
investment approaches in equity hedge funds.68 For the firm size effect see the work of Banz (1981) and Reinganum (1981).69 See for example Rosenberg et al. (1985).70 See Blatter (2003), p. 12.
3 Hedge Fund Strategies 38
securities.71 Tradeable market factors are utilized in quantitative multifactor models
in order to determine optimal long/short portfolios that generate attractive returns,
usually within a medium-term holding period.
As a third approach to portfolio composition statistical arbitrage strategies with
equities are based on the expected persistence of historical relationships (technical
analysis) or in general on the predictability of movements between equities or groups
of equities.72 While hedge fund managers with fundamental approaches are basically
stock pickers, the statistical arbitrage is model based.73 These strategies make use of
statistical techniques to identify and exploit temporary pricing imbalances that are
very often supply- or demand-driven.74 Common tools in this discipline are classical
time series techniques, including for example cointegration models, bayesian models
with a priori information, autoregressive models, vector error correction models,
pattern recognition or statistical decision theory. The trading frequency with these
strategies is very high and the investment horizon is usually short term.75
Often hedge funds traders combine the approaches outlined above. A manager could
for example use a top-down approach to identify a market and sector focus. Then
the fund might employ fundamental bottom-up techniques to pick out the most
promising stocks within the chosen market or sector. Finally statistical or technical
analysis could be applied to time entry and exit points.
3.2.1.3 Equity Hedge Trades
Classical statistical arbitrage trades are pair trades with long and short positions in
two related shares of a particular industry or sector.76 The objective in a pair trade
71 See for example Ineichen (2000), pp. 49-50.72 See Parnell (2001b), p. 63. Sometimes statistical arbitrage with equities is considered as an
individual hedge fund strategy, see for example the HFR index definitions.73 See Jaeger (2002), p. 54, and Ineichen (2001), p. 7.74 See Adrian (2003), p. 3, and Blatter (2003), p. 10.75 See Blatter (2003), pp. 10-11.
3 Hedge Fund Strategies 39
is to profit from one stock outperforming another. These stocks are often identified
with mean-reversion models where their current market price is compared to the
historical development or trend.77 Therefore, the hedge fund attempts to profit
from the likelihood that prices will trend toward a historical norm. So the hedge
fund manager has a positive expectation for one of these stocks which is held long in
the portfolio and a negative or less positive forecast for the other company resulting
in a corresponding short position. Realizing the resulting long/short spread is the
objective of the hedge fund’s bet. To avoid market and even industry risk, the two
companies involved in this trade are often competitors within the same industry.78
In statistical index arbitrage an index, exchange traded fund or index future is
traded against constituent stocks. The replication of broad market indices (usually
plus a spread) with few stocks is a very suitable hedge fund strategy for statistical
cointegration models79 as the cointegration analysis aims at minimizing the track-
ing error and maximizing the stationarity of the resulting (net) time series. This
stationary time series results from the difference of two integrated time series: index
(plus spread) and replicating portfolio. The profit is generated from superior returns
with less volatility and a smaller turnover than with other replication strategies.80
Intersectoral arbitrage is another field for statistical analysis. With econometric ap-
proaches the influence of key economic indicators on cross-sector spreads is analyzed
and forecasts are developed. Depending on the modeling results the spread is judged
76 For different assessments or categorizations of the terms “pair trading” and “statistical arbi-
trage” see Campell et al. (1999), p. 16, and Ineichen (2001), pp. 6-7.77 The identification of the components of a pair trade may also be done by fundamental methods,
see for example Kao (2002), p. 33, or Ineichen (2001), p. 6.78 See Nicholas (2000), p. 228.79 Cointegration in financial markets refers to co-movements in asset prices instead of co-
movements in returns like in other models. The econometric toolbox offers cointegration analysis
to detect common trends in various time series of stock prices.80 For the application of cointegration analysis in equity hedge strategies see for example Alexander
et al. (2001), Ionescu (2002), or Rehkugler/Jandura (2002).
3 Hedge Fund Strategies 40
as overvalued or undervalued. The resulting investment in the spread is built up
with long and short equity positions from these sectors.
trade resulting positions
long/short portfolio long/short over- and undervalued securities
pair trades long/short positions in two related stocks
index arbitrage long/short indices and constituent stocks
intersectoral arbitrage cross sectoral long/short positions
Table 3.1: Typical equity hedge trades81
The desired positions for the different equity hedge trades in Table 3.1 are usually
established with cash instruments (stocks), with derivatives like futures and options
or with over-the-counter contracts.82
3.2.1.4 Equity Hedge Risk Factors
Various risk factors are associated with equity hedge investing (see Table 3.2 on the
following page). An important source of hedge fund returns with equity strategies
is market risk. Due to positions in different stocks the portfolio is exposed to the
specific risk of these companies but the manager can also take risk on a sector or
market level.83 This stock picking risk includes for example changing profit margins
or market shares, regulatory issues, research and development and management
ability.84
Equity market risk is usually hedged on a revolving basis in equity market neutral
81 See Ineichen (2000), pp. 31-32 and 49-50, Cottier (1998), pp. 125-126, and Nicholas (2000),
pp. 219-247.82 See for example Cottier (1998), pp. 125-128.83 See Ahmad (2003), p. 12.84 See Nicholas (2000), p. 227.
3 Hedge Fund Strategies 41
strategies but when the long and short position do not match perfectly as with
long/short strategies the portfolio is also exposed to systematic market movements.85
For the short leg of long/short trades the risk factors in Section 3.2.8 on short
selling have to be considered. Liquidity risk may arise when the hedge fund is
engaged in stocks with a small capitalization. When long (or short) positions in
small capitalization stocks shall be sold (or bought back) in the market, there might
be liquidity problems.86 These capacity constraints arise especially if the size of
the position is large, compared to the company’s overall market capitalization.87 A
breakdown of long term relationships or patterns can cause serious problems with
the statistical methods. As statistical arbitrage techniques usually try to indicate
future convergence or divergence based on the analysis of historic movements this
is a source of model risk. Extraordinary factors or a less liquid market environment
can have a severe impact on the relationship of the financial instruments under
consideration. Another source of model risk is the fundamental method employed
in the valuation of individual companies. Table 3.2 sums up the central risk factors
in equity hedge strategies.
risk factor description
market risk - equities company specific price risk
systematic net long/short bias in the portfolio
liquidity risk market squeezes
model risk model based detection of opportunities
Table 3.2: Important risk factors in equity hedge strategies88
When compared to traditional long-only funds, the equity hedge portfolio is slightly
85 See Lhabitant (2002), p. 81.86 Primarily these capacity problems are encountered on the short side, see Tisdale (2000), p. 172.87 See Nicholas (2000), pp. 228-229.88 See Lhabitant (2002), p. 81, Smith (2001), p. 3, and Nicholas (2000), pp. 227-229.
3 Hedge Fund Strategies 42
more concentrated. The focus of single trades or parts of the portfolio can be
on regions, sectors, market capitalization and/or other exposures but the (equity)
hedge fund portfolio is usually diversified across a number of these market segments.
Leverage is often deployed to some amount but is used altogether scantily. The main
performance variable in equity hedge investing is the stock picking ability of the
hedge fund manager on the long and on the short side. Size is another key variable
for the success of equity hedge funds or strategies. When the amount of assets under
management is too large, the performance of market neutral equity hedge strategies
might suffer.89
3.2.2 Convertible Arbitrage
The strategy group convertible arbitrage manages about five percent of the capital
invested in hedge funds. Convertible arbitrage managers try to realize profits from
positions in mispriced convertible securities. Convertible securities are fixed income
instruments90 and they are usually issued as convertible bonds or as convertible
preferred stock91 which are exchangeable into common stock of the issuer or another
company.92 Another category of convertible securities are mandatory convertibles
which must be converted at the issuer’s option.93 For the issuer, convertibles are
a funding vehicle with a minor impact on cash flows and also a strategic tool in
mergers and acquisitions.94
As the holder of a convertible security has the right to exchange a fixed income se-
89 See Ineichen (2000), p. 49, Nicholas (2000), pp. 229-230, and Parnell (2001b), p. 65.90 Therefore, convertible arbitrage is sometimes categorized among the fixed income arbitrage
strategies, see Section 3.2.3.91 Convertible preferred stock represents equity rather than debt and is therefore subordinated to
debt of the issuing company.92 See for example Tremont (2000a), pp. 10-11.93 The following analysis is primarily focused on convertible bonds. In the US convertible bonds
made up about 50 percent of the convertibles market, see Tremont (2000a), p. 8.94 See Ahn/Wilmott (2003), p. 55.
3 Hedge Fund Strategies 43
curity for a fixed number of shares, the value of a convertible bond can be notionally
divided into a fixed income component and a so-called equity kicker containing the
optionality’s value.95 The fair value of such hybrid instruments has to be deter-
mined with pricing models, so the modeling assumptions and the required estimates
are potential sources of subjective views or errors. These models have to consider
the price process of the underlying security and the development of interest rates.
Consequently, hedge fund managers try to isolate individual risk premia and exploit
pricing inefficiencies between convertible bonds and related shares, bonds and/or
derivative securities.96
Convertible bonds are often undervalued compared to their theoretical value.97,98
Especially two reasons may account for this gap in primary offerings as well as in
the secondary market: the analysts’ coverage of convertible bonds and the issuers
rating. As many convertible bond issues are small in size only few analysts follow
them and therefore the transparency in this market segment is limited.99 The second
reason can be found in the non-investment grade rating of many convertible bond
issuers. This makes them suitable for a limited set of (institutional) investors causing
a lower liquidity in such convertibles.100
The most intuitive strategy with these undervalued securities is to buy and hold
95 Besides this (usually American type) call option convertible securities very often contain a
put option as they are callable, allowing the issuer to force the conversion into equity, see for
example Lhabitant (2002), pp. 84-85.96 See for example Ineichen (2000), p. 24 or Lhabitant (2002), p. 84.97 See Ammann et al. (2003) for a detailed analysis of the French market for convertibles. In this
study the theoretical values for the analyzed convertibles are on average 3 percent higher than
their observed market prices. The longer the time to maturity the more convertible bonds tend
to be underpriced and the market also significantly underprices convertibles that are at-the-
money and out-of-the-money, see Ammann et al. (2003), pp. 648-651.98 This theoretical value is usually based on market and credit risk, but does not take liquidity
risk into account.99 See Lhabitant (2002), p. 86.100 See Safaty (2001), p. 114, and Yee (2004), p. 21, for an overview of investment grade and
non-investment grade convertible bond issues.
3 Hedge Fund Strategies 44
them until their conversion, carrying various market and credit risks. Hedge funds
usually realize more sophisticated strategies (see Table 3.3 on page 47 for various
hedge fund strategies). A first convertible arbitrage hedge fund strategy is to hold
convertible bonds while hedging equity market risk with short positions in the un-
derlying stocks.101 When such a position is delta neutral102 the portfolio profits from
movements in the underlying stock’s market price. Thus the portfolio is long equity
volatility.103 Returns with these so-called cash-flow trades104 are generated from the
convertible bond’s coupon, the rebate on the short position in the underlying stock
and the delta neutral trading strategy.
The term convergence trading with convertibles aims at the detection of under-
or overpriced convertible securities. The long or short position in the convertible
is hedged synthetically with equity and interest rate instruments. Over time the
values of the hedge portfolio and the convertible security should converge and the
manager might be able to lock in a profit from the pricing discrepancies.105 This
profit is simply the difference between the original convertible value and the hedging
cost.106
But often even the delta hedge is left incomplete, for example when fewer stocks
are sold short in the beginning or the adjustments after a change in market prices
are missed out. The result is a net long position in the underlying stock which is
held in the portfolio exposing the strategy to additional equity risk.107 This so-
called bullish hedge profits from increasing prices in the underlying stock. A bearish
101 See Lhabitant (2002), pp. 86-88.102 Delta refers to the sensitivity of the convertible bond’s price to changes in the underlying stock.103 With the delta hedging of convertibles a straddle-like portfolio (long gamma) is created, see for
example Parnell (2001a), p. 55, or Calamos (2003).104 See for example Yee (2004), pp. 22-23.105 See for example Yee (2004), p. 25.106 These trades conform with the strategy label “convertible arbitrage” as a real arbitrage is
focused.107 See Ineichen (2000), p. 25.
3 Hedge Fund Strategies 45
hedge leaves a negative delta in the portfolio and is set up by selling short more
underlying stocks than for a delta neutral position. Therefore, a decreasing stock
price is a source of profit for this alteration of the classical delta (neutral) hedge. In
these at least partially delta hedged portfolio constructions interest rate and credit
risk is still carried by the hedge fund.108
Distressed trading with convertible securities focuses on convertibles issued by com-
panies that are threatened by financial distress. This is a more event-oriented trading
strategy as the manager bets on the restructuring process of the target company.109
Here the convertible bond’s price does not behave like a common convertible (which
is highly dependent on the underlying equity) because the inherent call option is
usually deep out-of-the-money.110 When setting up the trade, the distressed con-
vertible is bought and the equity market risk is hedged with a short position in
the underlying stock.111 Here the manager sometimes takes a larger short position
betting on a further depreciation of the stock or the bankruptcy of the company (in
analogy to a bearish hedge).112
In a Convertible Bond Asset Swap (CBAS), different arbitrage trades are com-
bined.113 The basic elements of such a hedging transaction can be summarized as
follows.114 The hedge fund manager acquires a convertible bond in the market and
sells this security to a so-called credit investor, usually an investment bank. In ex-
change he receives the value of the convertible’s bond floor and an OTC American
108 To hedge these risk factors, credit and interest rate derivatives can be used. Hedging the interest
rate risk can be easily done with interest rate derivatives like futures, forwards or swaps while
the credit risk component which arises when convertible bonds are unsecured or subordinated,
can be hedged with derivatives like credit default swaps, see Yee (2004), p. 25.109 See Section 3.2.6 on page 64 for the related event-driven strategy “distressed securities”.110 See Tremont (2000a), p. 13.111 See Ineichen (2000), p. 25.112 See Yee (2004), p. 25.113 Such asset swaps have become a industry standard in hedging risk, see for example Blatter
(2002a), p. 12.114 As the agreements can be very complex, the basic structure is open to numerous variations.
3 Hedge Fund Strategies 46
type call option on the convertible bond. The strike price of this option is the bond’s
present value calculated with a (recall) credit spread that is slightly smaller than the
spread for the evaluation of the bond floor. The resulting position is still exposed
to equity market risk that can be delta hedged by the manager. Credit and interest
rate risk is transferred to the credit investor.115 The hedge fund’s net position after
a delta (neutral) hedge is a long position in the option on the convertible bond
(which is identical to the (cheap) option component in the convertible bond) while
the downside potential is limited to the option premium. This strategy will profit
from tightening credit spreads as the hedge fund will exercise his call option on the
convertible bond and establish a new asset swap at a higher bond floor.116 Table 3.3
on the following page sums up some of the more important convertible arbitrage
trades.
The risk factors in different strategies of convertible arbitrage investing are listed in
Table 3.4 on page 48. They are straightforward when the positions resulting from
the trading strategies mentioned above are examined in detail. For some trading
strategies equity market risk is associated with the net position in the portfolio.
The very common long bias in the equity market creates a positive exposure while
a net short delta position has a negative correlation with this risk factor. Further
an at least partly delta neutral strategy is short (realized) equity volatility as the
resulting profit depends on the actual hedging strategy and transaction costs on the
underlying.117 As the value of the conversion option in the portfolio is associated
with the anticipated implied volatility (reflected in the options market price and also
incorporating other market frictions), the portfolio NAV increases with the value of
the option component.118
115 With the credit swap transaction the trade is exposed to counterparty risk which could be
severe in a credit crisis like the Russian debt default in 1998, see Tremont (2000a), p. 16.116 See Lhabitant (2002), pp. 88-90.117 If the delta hedging is done with matching standard options, this problem might be less severe.118 See Ineichen (2000), p. 24.119 See Ineichen (2000), p. 25, Yee (2004), pp. 22-25, Cottier (1998), p. 133, and Nicholas (2000),
3 Hedge Fund Strategies 47
trade positions sources of profit
delta hedging long convertible bond, coupon,
short underlying stock, short rebate,
long implied volatility, stock price movements,
short equity volatility conversion premium
bullish/bearish additionally long/short increasing or
hedging equity market decreasing stock price
convergence fully hedged convertible pricing discrepancies
trading
CBAS long convertible volatility, tightening credit spreads,
short underlying stock stock price movements,
conversion premium
Table 3.3: Typical convertible arbitrage trades and sources of profit119
If the bond component is held in the hedge fund’s portfolio, interest rate, market
risk and credit risk influence the success of the convertible arbitrage strategy.120
This is not the case for CBASs where credit and interest rate exposure is avoided
or exchanged into counterparty risk.
Convertible arbitrage is especially dependent on valuation models. The detection of
opportunities and the hedging of the very often complex conversion features is done
employing option pricing models121 and is therefore subject to model risk.122
pp. 192-194.120 See Parnell (2001a), p. 55. For the credit quality of convertible bond issuers in different markets,
see Safaty (2001), pp. 114-116.
3 Hedge Fund Strategies 48
risk factor description/position
market risk - equities long (or short) underlying company,
long implied volatility,
short realized volatility
market risk - interest rates long convertible bond
credit risk long (or short) underlying company
model risk model based detection
of opportunities and hedging
liquidity risk large bid/ask in convertibles,
short squeeze in underlying
event risk regulatory and prospectus risk
Table 3.4: Important risk factors in convertible arbitrage hedge funds123
Often, the pricing of the convertible portfolio components also relies on these models
which gives rise to another form of model risk, the mark-to-model risk.124 The
hedging process in most convertible arbitrage strategies typically requires short sales
in order to adjust the portfolio’s delta. A common risk factor from short selling is a
short squeeze which forces the manager to buy back shares at inflated prices. The
illiquidity of the convertibles market is a chance for arbitrageurs and funds willing
to capture this risk premium but on the other hand it is also a source of risk.125
121 See for example Ayache et al. (2002) or Wilmott (2000b), pp. 597-610, for convertible bond
pricing models.122 See Cottier (1998), p. 133.123 See Ineichen (2000), pp. 24-25, Cottier (1998), p. 133, Lhabitant (2002), p. 91, Parnell (2001a),
p. 55, Smith (2001), p. 3, and Nicholas (2000), pp. 194-198.124 In cases where a liquid option market on the relevant underlying exists this point is usually less
severe. The prices or implied volatilities of standardized option contracts can serve as proxy
for the conversion option. But as the features of plain vanilla options differ significantly from
the exotic construction of most convertibles, these modeling inputs have to be treated very
carefully.
3 Hedge Fund Strategies 49
Legal risk may arise due to intransparent or complicated details in the convertible’s
prospectus (for example concerning the treatment of dividends) or as a consequence
of regulatory intervention.126
Diversification in a convertible arbitrage hedge fund’s portfolio is usually achieved
by careful consideration of exposures to factors like industry, sector, credit quality,
implied volatility and event risk.127 The use of leverage in most cases ranges from 2
to 10 times equity.128 A key factor for convertible arbitrage hedge fund’s profits is
the supply with cheap convertible bonds although historically bullish hedges (with
a net long position in the equity market) were a source of substantial gains, too.129
3.2.3 Fixed Income Arbitrage
In this section we focus on hedge fund arbitrage strategies with interest paying
instruments (or their derivatives). There is a variety of investment strategies that
can be summarized under this broad category. According to HFR numbers a total
of five percent of assets under management is invested in this strategy group. The
core of these trades can be described as buying cheap fixed income instruments
and selling short (relatively) expensive securities. In the following we focus on the
bond/old-bond spread, basis trading, yield curve arbitrage, bond options trading
and spread arbitrage.130
A traditional hedge fund trade is the on-the-run and off-the-run bond arbitrage.
This issuance-driven trade holds a long position in an “old” bond from a recent
125 See Parnell (2001a), p. 55, and Nicholas (2000), p. 197.126 See Ineichen (2000), p. 24.127 See Nicholas (2000), pp. 196-197, and Lhabitant (2002), p. 91.128 See for example Ineichen (2000), p. 25, Parnell (2001a), p. 55, and Cottier (1998), p. 133.129 See Ineichen (2000), p. 25, and Peetz/Compton (2003), p. 203.130 The strategies convertible arbitrage and mortgage backed securities arbitrage (see Section 3.2.2
on page 42 and Section 3.2.4 on page 55) are sometimes classified as fixed income arbitrage
strategies as they involve interest bearing instruments.
3 Hedge Fund Strategies 50
issuance and a short position in the corresponding “new” bond from the current
auction or issuance.131,132 The reason for this set-up is the striking characteristic
that very often a significant premium is attached to the new bond.133 As time
passes the spread between the old and the new bond will converge toward zero.
Therefore, shorting the expensive new bond134, earning the repo rate on the sale
proceeds and purchasing the old bond in order to lock in the position on the spread
has the potential to generate trading profits. Usually this spread position is held
until the next auction date and then the position is unwind at a smaller spread.135
Basis trades are classical arbitrage transactions. Usually futures contracts on (gov-
ernment) bonds are sold and a bond that fits the delivery clause is bought simulta-
neously. As it is not sure which bond will be the cheapest at the delivery date and if
the supply in this cheapest-to-deliver (CtD) bond will meet the amount demanded
for the expiring futures contracts, there are various opportunities for arbitrage prof-
its. A net profit can be locked in when the bought bond’s price (including financing
costs and accrued interest) is smaller than the sold futures contract.136 This trade
sounds simple but requires very sophisticated tools when all the delivery-related
options in the futures contract are considered or valued.137
Yield-curve arbitrage focuses on mispriced yield differences for various maturities.
This group consists of trades like arbitrage in different maturities or the anticipation
131 See Ineichen (2000), p. 28.132 The spread between on-the-run and off-the-run bonds is used as explanatory variable in numer-
ous multifactor models for hedge fund returns, see for example Edwards/Caglayan (2001b), pp.
1009-1010.133 For a recent analysis of the bond/old-bond spread with the 30-year Treasury bond see
Krishnamurthy (2002), pp. 464-465.134 In the case of the 30-year Treasury these bonds are usually highly liquid, see for example Beder
et al. (1998), p. 296.135 See Krishnamurthy (2002), pp. 472-478, for an in-depth analysis of profits from this strategy
with 30-year Treasury bonds over the period 1995 to 1999.136 See for example Nicholas (2000), pp. 209-210.137 See for example Wong (1993), pp. 196-202, for the different option features.
3 Hedge Fund Strategies 51
of yield curve movements. When single bonds or maturities show mispricings relative
to the yield curve (graphically the yield curve might show kinks) the arbitrage trade
is usually realized with very close maturities. An assumption with most trades is
that the shape of the yield-curve remains unchanged for the duration of the trade.138
An example for yield-curve arbitrage is a butterfly trade where a bond is identified
as expensive relative to other bonds. Shorting the too expensive bond would create
a high amount of interest rate market risk. A butterfly trade minimizes the risk of
changes in the yield curve as there are two “wings” - one trade with a longer and one
trade with a shorter maturity against the mispriced security. Therefore, one long
position will profit from a steeper yield curve while the other long position bets on a
flattening yield curve.139 Trading directly on yield curve movements (with so-called
yield curve shape trades) is much more risky than this. When a hedge fund manager
bets on a decreasing or increasing slope of the yield curve, there is usually no effective
hedge in place. One bond is held long and another bond with a significantly different
maturity is sold short. Profits are made when the forecast, which is often conducted
with the assistance of statistical analysis (to isolate seasonalities or irregularities),
holds true.140
Bond options (or options on bond futures) are a further playing field for hedge fund
managers. Arbitrage strategies include the exploitation of bond option mispricings
via put-call-parity, the purchase (sale) of cheap (expensive) options, combined with
delta hedging or the trading of implied volatilities of options.141 Embedded option-
like features in bonds are another field that is open to arbitrage trading. If these
implicit bond options (for example call features) are mispriced, they can be isolated
and picked up by the arbitrageur.
Spread arbitrage concentrates on differences in the yields of various bonds. The
138 See Wong (1993), p. 226.139 See for example Tuckman (2002), p. 77-82.140 See for example Wong (1993), p. 206.141 See Wong (1991), p. 212, and Wong (1993), pp. 211-225.
3 Hedge Fund Strategies 52
most common spread is observable between corporate and government bonds. As
this bond spread can be partially attributed to different influences like the corpo-
rate issuer’s credit risk, taxes on corporate bonds and systematic risks142 (further
influences might be a specific bond’s liquidity or call features), the trader’s sub-
jective views about the development of these factors decide about the positions he
should enter. Expecting a widening spread, the hedge fund will enter a long po-
sition in government bonds or bond-futures and sell short the respective corporate
bond. The spread arbitrage can also take the form of a relative-value trade when the
trade is established with two different corporate bonds.143 The term TED-spread
(Treasury/Eurodollar-spread) originally referred to the difference between yields on
US Treasury Bills and yields on Eurodollars (certificates of deposit in US dollars in
a non-US bank).144 Nowadays the TED-spread often stands for general positions
in (global) government bonds and par swaps.145 The TED-trades are usually con-
structed by trading Treasury bill (or note) futures against Eurodollar futures.146
When the spread is expected to widen, Treasury bill futures are bought and the
equivalent amount of Eurodollar futures is sold. Asset swap spread trades with
corporate bonds are very popular nowadays. In such an asset swap transaction a
bond is purchased and the bond’s fixed-rate cash flows are immediately exchanged
for floating-rate cash flows with an interest rate swap. If the floating rate is above
the bond’s financing costs, there is a potential profit in this trade. The floating-rate
will bear some spread over a (risk-free) reference rate, for example LIBOR.147 This
spread is the reward for (or a function of) credit risk inherent in the bond position.
142 See the study of Elton et al. (2001) for details on these influences.143 See for example Lhabitant (2002), p. 93.144 So this spread reflects views of the relative credit quality of the US Treasury and high quality
international financial institutions, see Lhabitant (2002), p. 94.145 See Nicholas (2000), p. 211.146 See CBOT (2002) for an implementation with interest rate swap futures. Kawaller (1997)
provides further information on the implementation of corresponding trades.147 See for example Hull (2003), pp. 618-619.
3 Hedge Fund Strategies 53
trade resulting positions
on- and off-the-run trade long/short two close maturities
basis trade long CtD bond, short future
butterfly trade long/short three close maturities
yield curve movements long/short different maturities
bond option trading long/short (implicit) options and underlying
spread arbitrage long/short different credit qualities
bond asset swap long bond, interest rate swap
Table 3.5: Typical fixed income arbitrage trades148
The introduced fixed income trades are listed in Table 3.5.
Interest rate risk is the central market risk for this basket of fixed income trades.
Usually it is hedged in fixed income arbitrage strategies, but when yield curve arbi-
trage trades are considered there is no extensive hedge in place. The main compo-
nents of this risk factor are parallel shifts in the yield curve, steepening or flattening
of the yield curve, increased or decreased curvature of the yield curve and the volatil-
ity of interest rates.149 As second market risk residual currency exposure has to be
examined when securities denominated in different currencies are arbitraged. They
are usually hedged with currency futures contracts, but in practice these hedges
are not always perfect.150 Another risk factor is credit risk. Especially in spread
arbitrage trades this is the central risk factor. Differences in default risk cause the
relevant spreads, and changes in the estimated ability to repay interest and principal
are responsible for increasing or decreasing spreads.151 Liquidity risk arises in two
148 See Wong (1993), pp. 196-226, Ineichen (2000), pp. 28-29, Cottier (1998), pp. 128-130, and
Nicholas (2000), pp. 213-215.149 See for example Tremont (1999), p. 5.150 See Nicholas (2000), p. 214.151 See Nicholas (2000), p. 214.
3 Hedge Fund Strategies 54
forms. On the one hand the success of most strategies or trades is contingent on low
net financing costs, especially when short positions are set-up. On the other hand
fixed income arbitrageurs usually hold less liquid bonds as they trade at a discount
and can therefore be arbitraged against liquid securities trading at a premium.152
The usage of quantitative modeling is very pronounced in fixed income arbitrage
hedge funds. Mispricings are detected with sophisticated models and apart from
true arbitrage trades, the success of trades relies on realistic modeling assumptions
and accurate predictions.153 Event risk with fixed income instruments includes gov-
ernment policy moves like the issuance of new bonds or changes in tax laws. But
both risk factors can also open up new opportunities for hedge funds. The different
sources of the risk premia for fixed income arbitrage hedge funds are summarized in
Table 3.6 on the next page.
The margins in fixed income arbitrage are usually very small in relative numbers,155
so the positions have to be levered in order to offer attractive returns.156 The
leverage is achieved through borrowing, repurchase transactions or derivatives. This
extensive leverage ranges from 20 to 30 times equity157 (and the fund volume can
also reach values of 150 times equity158) confirming the importance of attractive
financing conditions in fixed income investing.
152 See Ineichen (2000), p. 28.153 See Nicholas (2000), p. 214.154 See Ineichen (2000), pp. 24-25, Cottier (1998), p. 129, Smith (2001), p. 3, and Nicholas (2000),
pp. 209-218.155 Ineichen (2000) talks about 3 to 20 basis points, see Ineichen (2000), p. 29.156 See Wong (1993), pp. 239-240.157 See Ineichen (2000), p. 29.158 See Cottier (1998), p. 129.
3 Hedge Fund Strategies 55
risk factor description/position
market risk - interest rates usually hedged,
existent in yield curve arbitrage
market risk - currencies residual exposure to different currencies
credit risk long/short default risk,
short positions usually in better ratings
liquidity risk low financing costs,
long positions in illiquid instruments
event risk issuance of new government bonds,
changes in tax laws, etc.
model risk model based detection
of opportunities and hedging
Table 3.6: Important risk factors in fixed income arbitrage hedge funds154
3.2.4 Mortgage-Backed Securities Arbitrage
Mortgage-backed securities (MBS) arbitrage hedge funds invest in mortgage-backed
securities and corresponding derivatives.159 MBS differ from regular bonds with
respect to their cash flow structure, the amortization of principal and prepayment
possibilities.160 Because of such special features, this is an interesting area for ar-
bitrageurs.161 The US MBS market with an outstanding amount of more than 4
trillion dollars162 is especially attractive due to various factors. In the first quarter
of 2005 MBS hedge funds reflect nearly three percent (HFR estimate) of the global
hedge fund capital. Positive influences are for example the governmental (or quasi-
159 See Cottier (1998), p. 131.160 See Fulenwider (1996), p. 122.161 The markets for MBS are fully developed in the US and in the UK, see Cottier (1998), p. 131.162 See Perli/Sack (2003), p. 8.
3 Hedge Fund Strategies 56
governmental) standardization of the issues and the therefore high credit quality,
the tax relief for mortgage interest and the availability of consumer credit informa-
tion.163 Like convertible arbitrage (see Section 3.2.2 on page 42), this strategy is
sometimes classified as a subcategory of fixed income investing (see Section 3.2.3 on
page 49) although their cash flows depend on a broad range of influences.164
There exists a variety of mortgage-backed securities. The starting point of all
these structures are loans on residential and/or commercial mortgages. Loans
with similar characteristics are pooled into asset-backed securities (ABS) with pass-
through structures. The pooling and the issue of residential MBS is usually done
by government-sponsored organizations like the Federal National Mortgage Associ-
ation (FNMA) and the Federal Home Loan Mortgage Corporation (FHLMC), or by
government agencies like the Government National Mortgage Association (GNMA)
in the US and The Mortgage Corporation (TMC) in the UK.165 Numerous pri-
vate institutions such as investment banks or home builders also issue residential
and commercial mortgage-backed securities, which are referred to as non-agency or
private label MBS.166 Most MBS are of the highest rating class and sometimes in
addition guarantee capital.167 Pooling these MBS and/or original loans in so-called
collateralized mortgage obligations (CMO) creates a derivative structure that is
sold to investors in different tranches or classes with different cash flow characteris-
tics.168 These tranches carry individual risk profiles and receive different portions of
the interest and principal (so-called sequential-pay structure). There are numerous
variations of these CMO structures.
163 See Tremont (1999), p. 3.164 See for example Fung/Hsieh (2002b) for this categorization.165 See Wong (1993), pp. 38-39 or Nicholas (2002), pp. 131-132 for a brief description of these
organizations.166 See for example Friedland (2001).167 See Cottier (1998), p. 131.168 See for example Fulenwider (1996), p. 123, or Wong (1993), pp. 41-42.
3 Hedge Fund Strategies 57
The coupon rates of CMO floaters are periodically adjusted to some underlying index
(usually Libor). Super floaters adjust to a multiple of an index while the coupon of
inverse floaters reacts converse to the reference index. As the prepayment risk is high
with the underlying loans,169 investors are attracted by planned amortization class
(PAC) bonds offering significant prepayment protection. This protection is achieved
when other tranches or classes in the ABS structure absorb all the prepayment cash
flows and thus all the prepayment risk.170 Other CMO variations are the very
common interest-only (IO) and principal-only (PO) bonds. Here the payments in
the ABS collateral are stripped into interest and principal cash flows. Therefore,
they can serve as natural hedging instruments for mortgage portfolios.171 A Z-Bond
is a tranche of the CMO that receives no interest or principal payments until the
preceding classes are fully paid. So this type of CMO does not offer regular coupon
and it is sold at a discount (similar to zero coupon bonds).172 Inverse IO securities
combine the features of floating securities with the IO characteristic. Therefore, the
inverse IO coupon rises with decreasing interest rates and the average life and cash
flow falls with increasing prepayments.173
There are various possibilities for trades with MBS. Directional investment in MBS
tries to exploit views on the development of interest rates and/or the amount of
prepayments with long positions in different MBS instruments (for example with
inverse floaters).174 Market neutral or hedged strategies with MBS are very common
among hedge funds. They capture the spread between different MBS or CMO
structures or the spread between MBS and a reference like for example Libor or the
yield on government bonds. For some of these trades and the resulting positions see
169 See page 58 for more details on prepayment risk.170 See Friedland (2001).171 See Wong (1993), p. 43.172 See Nicholas (2002), p. 136, and Friedland (2001).173 See Nicholas (2002), p. 141, and Fulenwider (1996), pp. 124-125.174 See Tremont (1999), p. 4.
3 Hedge Fund Strategies 58
trade resulting positions
Directional MBS positions in interest rates and/or prepayment risk
MBS hedging long/short different pass-through MBS
MBS arbitrage long MBS and short government bonds or derivatives
CMO arbitrage long/short CMO and pass-through MBS
or long PAC and short other MBS or CMOs
Inverse IO trading positions in inverse IO
Distressed MBS long/short distressed MBS
Table 3.7: Typical mortgage-backed securities arbitrage trades175
Table 3.7.
Risk factors in MBS arbitrage trades are listed in Table 3.8 on the next page. As
MBSs in general offer fixed income features, they are subject to interest rate risk.
The risk of prepayment for MBS can be further divided into general turnover risk,
mortgager’s cost changes and changes in the mortgagers’ alertness.176 The prepay-
ment feature of the underlying loans creates an inverse exposure to interest rates.
When interest rates decline, a substantial part of the homeowners or corporations
will start to refinance their mortgages causing prepayment cash flows in the collat-
eral pool of mortgage-related ABS. Credit risk in MBS depends on the respective
issuer. As most mortgage securities are arranged by government institutions and
offer guarantees of principal, they attain a credit quality similar to treasury secu-
rities.177 Other “private” placements may be more exposed to this risk factor.178
Liquidity risk arises with some securities in the MBS market. There are very liquid
175 See Wong (1993), pp. 206-209, Tremont (1999), pp. 4-5, Lhabitant (2002), p. 95, Fulenwider
(1996), p. 125, and Cottier (1998), p. 132.176 See Tremont (1999), p. 5.177 See Lhabitant (2002), p. 95.178 See Tremont (1999), p. 6.
3 Hedge Fund Strategies 59
segments that are comparable to the corporate bond market,179 but there are also
some CMO structures (like for example inverse IOs) that lack liquidity and offer
opportunities to arbitrageurs.180
A further crucial point to MBS investing is the tax legislation. If the deductability
of mortgage interest would be limited or abolished this would have a severe impact
on the MBS-market.181 Another major risk factor can be found in the modeling
of MBS. In the course of detecting opportunities, MBS hedge funds have to apply
sophisticated pricing models for example in order to embed prepayment features or
interest rate scenarios, which exposes them to significant model risk.182
risk factor description/position
market risk - interest rates long/short fixed income securities
credit risk depending on issuer and guarantees
liquidity risk more complicated CMOs lack liquidity
event risk prepayment risk and tax legislation
model risk complex pricing models are used
Table 3.8: Important risk factors in mortgage-backed securities arbitrage183
Because of these risk factors, MBS trade at a yield premium of up to one or two hun-
dred basis points above comparable government bonds.184 A central source of return
in MBS arbitrage is the complexity of the mortgage related securities.185 Another
main influence is the ability to predict the mortgager’s behavior, as irrational human
179 See Friedland (2001).180 See Fulenwider (1996), pp. 124-125.181 See Tremont (1999), p. 6.182 See for example Keiter (2000), pp. 217-219.183 See Tremont (1999), p. 5, Fulenwider (1996), pp. 124-125, Cottier (1998), p. 131, Keiter
(2000), pp. 220-229, and Friedland (2001).184 See for example Tremont (1999), p. 6, and Friedland (2001).185 See Fulenwider (1996), pp. 125-126.
3 Hedge Fund Strategies 60
decisions due to non-financial reasons can be a major performance factor.186 Further
contributions to the performance are the behavior of spreads or credit mispricings.
Leverage with MBS arbitrage is usually less than three times equity. The diver-
sification in MBS hedge funds is achieved with investments in numerous positions
across a variety of mortgage sectors and different MBS structures.187
3.2.5 Merger Arbitrage
Merger arbitrage hedge funds pursue classical convergence trades. They focus on
the stocks of companies which are or will be involved in an acquisition or merger
process.188 As there is some risk or probability that such a transaction will not
be completed, the target company’s securities trade at a discount, expressing the
market’s view about success or failure.189 This is the spread or risk premium merger
arbitrage hedge fund managers are interested in.190 The time horizon of merger
arbitrageurs is short term. Although the spreads in merger arbitrage are typically
narrow, the returns are significant on an annualized basis as the arbitrage trans-
actions usually can be realized in a short period of time.191 In 2005 the amount
invested in merger arbitrage hedge funds is about 1.5 percent of the net asset value
in the hedge fund universe.
In general there are two different types of mergers, stock mergers and cash mergers,
depending on what a bidder offers in exchange for the target companies equity.192
186 See Cottier (1998), pp. 131-132.187 See Tremont (1999), p. 7.188 Some definitions differentiate between merger and risk arbitrage. In Ineichen (2001) risk arbi-
trage includes merger arbitrage as well as special (corporate) situations, see Ineichen (2001), p.
24.189 See Cornelli/Li (2001) for the role of merger arbitrageurs in takeovers.190 See Yang/Branch (2001) for a detailed review of academic literature on merger arbitrage.191 See Lhabitant (2002), p. 112.192 More complicated situations may arise when warrants, preferred stocks, debentures or multiple
bids are involved, see for example Lhabitant (2002), p. 111.
3 Hedge Fund Strategies 61
In a cash merger193 (also called “cash tender offer”) a hedge fund is simply long the
stock of the company being acquired. This aims at realizing the spread between the
market price at which the stock currently trades and the tender price realized when
the deal is completed (the premium offered by the acquiring company). The dividend
paid during the holding period on the target company’s stock is also realized by the
arbitrageur.194
In a stock merger (also called “stock for stock merger” or “stock swap merger”195) an
arbitrageur is usually short the stock of the acquiring company and long the stock of
the acquisition target. This is more complicated than in a cash merger as the offer in
such a transaction is not fixed but depends on the bidding company’s stock price.196
When the spread between the stocks involved in this transaction disappears, a profit
from the long and short positions can be realized.197 Another source of profit can
be the dividends of the stocks held long, but this should be offset by the dividends
for the short position in the acquiring company.198 Finally interest on the short
sale proceeds is a source of profits in stock merger arbitrage.199 Table 3.9 on the
following page shows the positions in typical merger arbitrage trades.
As the success of a merger or acquisition depends on the approval by the shareholders
of the acquisition target, the regulation authorities (Departement of Justice, Fed-
eral Trade Commission, European Competition Bureau) and management actions,
193 A brief description of the history of cash mergers is given in chapter four of Wyser-Pratte
(1982).194 See Mitchell/Pulvino (2001), pp. 2138-2139.195 See for example Ineichen (2000), p. 34.196 An innovation in merger deals are so-called collar mergers as a special form of stock mergers.
Here the exchange rate (acquirer shares for target shares) is variable and depends on the ac-
quirer’a and/or the target’s stock price, see Yang/Branch (2001), p. 18, Hsieh (2001), p. 30,
Mitchell/Pulvino (2001), p. 2170, and Blatter (2002b), p. 19.197 See Ineichen (2000), p. 34.198 See Mitchell/Pulvino (2001), pp. 2138-2139.199 See Mitchell/Pulvino (2001), p. 2139.200 See Ineichen (2000), pp. 34-37, Nicholas (2000), pp. 201-204, and Lhabitant (2002), p. 112.
3 Hedge Fund Strategies 62
trade positions sources of profit
cash merger long target spread, dividends, (opportunity cost)
stock merger long target, spread, net dividends, short rebate
short bidder
Table 3.9: Typical merger arbitrage trades and sources of profit200
problems during the negotiations have to be taken into account. This transaction-
based risk is a source of returns for hedge fund managers as this uncertainty causes
the required spreads which will often fluctuate as the market changes its opinion on
the probability of consummation.201 Herein lies the fund managers’s opportunity. A
successful merger or acquisition process will force these spreads to disappear. But in
case a deal is called off, the spreads will quickly widen again. In such a situation the
long position will loose value (the premium will disappear) and the short position
has to be covered at higher prices as several arbitrageurs will try to cover their short
sales at the time the merger proposal is terminated.202 These losses will usually be
much greater than the potential gains if the deal succeeds.203 In the case of stock
mergers this event or breakup risk can be influenced by market risk. When equity
markets are falling, the target company’s shareholders may refuse their consent as
they would be compensated in shares.204 Therefore, the strategy is exposed to a
long position in equity volatility.205 In case of an unsuccessful takeover due to a
sharp decline in the price of the bidder’s stock the gain in the short position would
compensate for the losses in the target company’s equity. A hedge fund manager
must also pay attention to the target company’s specific risk as a decline in value
201 See Parnell (2000), p. 3.202 See for example Wyser-Pratte (1982), p. 14.203 See Mitchell/Pulvino (2001), p. 2135, or Paulson (2000), p. 189.204 See Lhabitant (2002), p. 112.205 See Ineichen (2000), p. 34.
3 Hedge Fund Strategies 63
might also cause a breakup of negotiations.
Besides this legal or breakup risk another source of uncertainty is a delay of the
transaction, which causes the annualized return (resulting from the fixed spread) to
drop significantly. Regarding the net position in the arbitrageurs portfolio, further
risks can be identified. In a cash tender offer, the hedge fund is usually net long the
acquisition’s target and therefore in case of an unsuccessful takeover also exposed to
market risk when the premium vanishes. In a successful takeover the market price
should equal the tender price thus eliminating market risk. A stock merger normally
results in an arbitrage portfolio with a small short bias. The long and short positions
are usually not cash-balanced and thus not completely market neutral. The invested
volume in the long position is a bit smaller than the short position as the values
(when the exchange-ratio is considered) are expected to converge. Therefore, the
stock merger arbitrage portfolio is usually short delta to a small extent.206 In order
to reduce transaction costs, some managers use options instead of buying stocks
which makes them sensitive to changes in (implied) volatility. The central risk
factors in merger arbitrage hedge funds are summarized in Table 3.10.207
risk factor description/position
market risk - equities positions in target (and bidder) company,
net short position (stock merger),
position in implied and/or realized volatility
event risk approval of regulators and/or shareholders
Table 3.10: Important risk factors in merger arbitrage hedge funds208
Merger arbitrage hedge funds diversify their portfolio by investing in several deals at
206 See Ineichen (2000), p. 34, or Drippe/Eyrick (2001), p. 67.207 See for example Paulson (2000) for more details on risk in Merger Arbitrage.208 See Ineichen (2000), pp. 34-37, Smith (2001), p. 3, Paulson (2000), and Lhabitant (2002), p.
112.
3 Hedge Fund Strategies 64
the same time.209 They also limit the size of capital allocation in individual deals and
often invest in transactions which involve different risk factors in order to have a low
correlation of success or failure in the portfolio of merger bets.210 The use of leverage
in merger arbitrage hedge funds is not standardized. Many managers employ at least
some amount of leverage depending on the arbitrage opportunities.211 An adequate
amount of mergers and acquisitions is naturally the core of the strategies in order to
build a diversified merger arbitrage portfolio. This quantity is a main performance
variable.212 In times of an economic upturn or a rising equity market, merger and
acquisition situations tend to occur very frequently.213 Another key factor in merger
arbitrage is a large or sufficient premium on successful transactions in order to
compensate for losses in unsuccessful takeovers. Further a declining interest rate
environment is a catalyst for this strategy.214
3.2.6 Distressed Securities
When following a distressed securities strategy, hedge funds invest in debt or equity
instruments of companies which face difficulties. Both, companies with financial
and/or operational problems, as well as enterprises which are already unable to fulfill
their liabilities or are in bankruptcy are counted among the distressed category.215
Therefore, the portfolio constituents of distressed securities investors predominantly
show ratings in the categories C (imminent default) and D (actual default).216
The spreads that attract hedge funds may result from absolute or relative pricing
209 See for example Ineichen (2000), p. 34, or Drippe/Eyrick (2001), p. 70.210 See Nye et al. (1996), p. 77.211 See for example Nicholas (2000), p. 205, or Drippe/Eyrick (2001), p. 69.212 See Lhabitant (2002), p. 112, and Ineichen (2000), p. 34.213 See Drippe/Eyrick (2001), pp. 67 and 70.214 See Drippe/Eyrick (2001), p. 70.215 See Ineichen (2000), p. 37.216 See Parnell/Matos (2001), p. 31.
3 Hedge Fund Strategies 65
inefficiencies.217 Absolute inefficiencies occur between a security’s fundamental or
intrinsic value and it’s market price while relative mispricings can come up between
the prices of two securities issued by the same company which is currently in dis-
tress.218
The market for securities of distressed companies is mainly a buyer’s market.219
Most individual investors are unwilling to bear the risks associated with securities
from companies in distress,220 and many institutional investors face legal restrictions,
which force them to sell bonds rated below investment grade.221 Finally the required
know-how for distressed securities investing, the time consuming monitoring process
of the distressed assets and the scarce coverage by analysts even convince holders
of distressed securities to sell below their fundamental value.222 In such a market
hedge funds that invest in distressed securities act as liquidity providers.
The strategies employed by distressed securities hedge funds are numerous, but
there are two main categories of distressed securities investing. Active hedge funds
try to gain significant influence as creditors in the restructuring and refinancing
process, while passive distressed securities managers buy “cheap” equity and/or
debt of distressed companies and simply hold until their investment pays off.223
When exploiting relative mispricings of distressed securities, i.e. inefficiencies be-
tween the prices of two securities issued by the same company, the trades are often
classified among capital structure arbitrage. During a restructuring process the cap-
217 See Lhabitant (2002), pp. 99-100, or Ineichen (2000), p. 37.218 This leads to so-called capital structure arbitrage, see for example Parnell/Matos (2001), p. 33.
Some convertible arbitrage trades (see Section 3.2.2 on page 42) could also be classified among
capital structure arbitrage as different instruments from one company are traded against each
other.219 See Cottier (1998), p. 134.220 See Ineichen (2000), p. 37.221 See Lhabitant (2002), p. 100.222 See Lhabitant (2002), p. 100.223 See Parnell/Matos (2001), p. 33, or Lhabitant (2002), p. 102.
3 Hedge Fund Strategies 66
ital structure arbitrage specialists purchase the undervalued security and take short
trading positions in the overpriced asset to generate an arbitrage profit.224 The
classical trade is to identify companies where the market value of equity is at a too
high level versus debt and to implement a trade consisting of a long debt and a short
equity position.
Distressed investing includes various instruments from secured debt to common
stock, for example debtor-in-possession loans, trade claims, secured bank debt, real-
estate loans, mortgages, senior or subordinated debt, letters of credit, (busted) con-
vertible bonds225, preferred or common stock.226 In these securities many hedge
funds very often only take long positions as it is difficult, risky or even impossible to
short distressed or even defaulted securities.227 Some typical trades with distressed
securities are summarized in Table 3.11 on the next page.
In the past, most investment opportunities for distressed securities hedge funds were
offered by US corporations. For the US Chapter 11 of the US Bankruptcy Code
provides a market oriented bankruptcy process which provides relief from creditor
claims for companies in financial distress.229 Depending on the filing, distressed
securities can be categorized in one of four stages of their life cycle. The pre-
bankruptcy period prior to the bankruptcy filing, the early-stage bankruptcy starts
immediately after the filing for bankruptcy protection and finally the middle- and
late-stages with due diligences and the distribution of new security baskets.230
Due to two major drawbacks, the European market has not been very attractive
224 See Ineichen (2000), p. 37.225 Busted means that the conversion feature is virtually worthless, see for example Parnell/Matos
(2001), p. 33.226 See for example Cottier (1998), p. 134.227 See Parnell/Matos (2001), p. 33.228 See Ineichen (2000), pp. 38-39, Cottier (1998), pp. 135-136, Parnell/Matos (2001), p. 33.229 If companies file for Chapter 7 of the US Bankruptcy Code they prepare for liquidation and
look for supervision in the liquidation process.230 See Tremont (2000b), pp. 7-8.
3 Hedge Fund Strategies 67
trade description
active debt/equity core positions influence on the restructuring process
long term passive debt/equity posi-
tions
buy and hold
investing in so-called orphan buy and hold
equities after reorganization
passive short term positions anticipation of a specific event
capital structure arbitrage different securities of one company
long term positions under buy and hold specific
partial hedging risk until reorganisation
Table 3.11: Typical distressed securities trades228
for distressed securities specialists.231 In Europe, under various jurisdictions, it is
harder to force a default as very often distressed securities are issued by holding
companies, whereas in the US the issuers are operating companies. A second reason
for the less liquid distressed securities market is the lack of transparency in most of
the national European bankruptcy laws.232
There is a variety of risk factors that affect distressed securities investing. Specific
risk of the distressed company (or companies) is the most obvious and essential
risk.233 Model risk is also a very important part of the overall risk profile. The search
for under- or misvalued securities the detection is usually model-based. As marking-
to-market is often impossible, net asset values are usually model-based. This may
result in lower (historical) risk figures due to performance smoothing.234 Equity
market risk of stock positions held long is to some extent hedged with short positions
231 See for example Paetzmann (2003) for problems with distressed investing in Germany.232 See Lhabitant (2002), p. 102.233 See for example Lhabitant (2002), p. 104.234 See Cottier (1998), p. 135.
3 Hedge Fund Strategies 68
in the distressed company’s sector or futures contracts, but cannot be fully avoided.
The strategy is usually long debt instruments with a low rating classification which
implicates credit risk.235 Liquidity risk accounts for another significant portion of
overall risk, as the hedge fund’s positions are generally illiquid. On the one hand,
after a restructuring, a controlling position cannot be sold immediately and on the
other hand, the road to the restructuring in itself may already take it’s time. This
leads to a strict redemption policy of distressed securities hedge funds with lock-
up periods of up to one year.236 The legal risk with distressed securities investing
comes from the often needed court decisions before the final restructuring. Positions
in distressed debt are usually long term and thus subject to interest rate risk, as
rising interest rates will decrease the strategy’s returns.237 A short review of the
risks associated with distressed securities investing is given in Table 3.12.
risk factor description/position
market risk long distressed company,
long duration position
model risk model based detection of opportunities,
model based NAV calculation
credit risk long default risk
liquidity risk illiquid positions, long lock-up periods
event risk court decisions
Table 3.12: Important risk factors in distressed securities hedge funds238
Diversification in order to reduce portfolio risk is achieved by investing in various
235 See Ineichen (2000), p. 37.236 See Lhabitant (2002), p. 104.237 See Ineichen (2000), p. 37, and Lhabitant (2002), p. 104.238 See Ineichen (2000), pp. 37-39, Cottier (1998), p. 135, Lhabitant (2002), p. 104, and
Parnell/Matos (2001), p. 37.
3 Hedge Fund Strategies 69
securities, companies and sectors. The leverage used with distressed securities in-
vesting is usually low.239 Key factors for success in distressed securities investing
are a superior ability to value a distressed company’s assets, negotiating and bar-
gaining skills and a thorough understanding of the relevant investment risks.240 As
economic downturns with deficient corporate returns often entail financial distress,
for many companies the economic environment is another key influencing factor on
distressed hedge funds.241 The market share of distressed securities hedge funds in
2005 is approximately five percent.
3.2.7 Global Macro
Macro or global macro hedge funds are very flexible in their investment approach
and try to profit wherever they identify opportunities. The investment process of
macro funds is the least restricted in the hedge fund industry.242 These funds usually
try to capture shifts in international monetary, political and economic policymaking,
which impacts interest rates and currency, stock and bond markets.243 Depending on
the macroeconomic environment managers follow completely different strategies.244
The strategies have a macroeconomic view in common and are applied in developed
and emerging markets.245 Very often macro hedge funds pursue a base strategy (for
example long/short equity) and take on different opportunities or bets depending
on the manager’s views.246 Due to spectacular trades, they were the hedge fund
industry’s biggest players and account(ed) for most media attention.247
239 See Cottier (1998), p. 135, Lhabitant (2002), p. 104, and Murray (2000), p. 239.240 See Tremont (2000b), p. 9.241 See for example Tremont (2000b), p. 24.242 See Ineichen (2003), p. 319.243 See DeBrouwer (2001), p. 21, and Lungarella (2004), p. 10.244 See for example Strome (1996), pp. 27-28.245 Immature markets like those for electricity or weather are new fields for macro hedge fund
managers, see Blackfish (2001) or Ahn et al. (2002).246 See Ineichen (2000), pp. 42-43.
3 Hedge Fund Strategies 70
The resulting strategies can be assigned to the three different classes directional
macro, long/short macro and macroeconomic arbitrage which are summarized in
Table 3.13 on page 72. Expectations on economic variables that influence the prices
of different assets are in the heart of directional macro trading.248 Positions (long
or short) in various assets are set up in order to profit from anticipated (global)
trends or movements in currencies, interest rates or emerging markets.249 These
bets on future divergences are the central strategy for global macro hedge funds.250
Here are many intersections with managed futures strategies (see Section 3.2.10)
as directional macro managers and managed futures traders often employ trend-
following models that are based on macroeconomic data.251
Long/short macro trading stands for simultaneous long and short positions in
different assets. Based on macroeconomic views, the direction or development of
such spreads is extrapolated. This strategy can be successfully implemented in
trending and non-trending markets, a feature that directional macro trading cannot
offer. So the long/short macro strategy is less correlated with the reference markets
than pure directional bets.252
In directional macro as well as in long/short macro trading, models are very impor-
tant tools to identify opportunities in different markets. Simple statistical mean-
reversion plays a central role in such macro models.253 Recent trading models apply
more sophisticated macroeconomic modeling approaches in order to detect disequi-
libria in financial markets.254 The decisions on these trades are usually discretionary
247 According to HFR data for the first quarter of 2005, 11 percent of the hedge fund capital is
invested in macro strategies. In the year 1990 HFR even reported a market share of 71 percent.248 See Burstein (1999), p. 14.249 See Cottier (1998), p. 140.250 Often literature on hedge funds mentions this strategy as the only macroeconomic approach,
see for example Lhabitant (2002), pp. 115-116.251 See Schmidt (2003), p. 24.252 See Burstein (1999), pp. 26-27.253 See Strome (1996), pp. 29-31.
3 Hedge Fund Strategies 71
but the timing of entry and exit dates is very often supported by technical analy-
sis.255 Examples for specific long/short macro trades can be found in the area of
currency options trading or in currency convergence trades.
The term macroeconomic arbitrage was introduced by Burstein (1999). Sub-
jective macroeconomic views are no longer the fundament for this group of trades.
Macroeconomic arbitrage is quantitative in nature and involves fundamental and
technical analysis.256 Macroeconomic arbitrage is defined as identification and ex-
ploitation of macroeconomic mispricings.257 Such a mispricing can be the divergence
of the market price ratio between different assets or groups of assets and a corre-
sponding relation of influential macroeconomic factors.258 For example, the ratio of
two sector indices of stocks and the ratio of two driving macroeconomic variables
(like the gross domestic products (GDP) of two countries) can track each other
for a long time.259 When a significant divergence in these ratios gets obvious, a
macroeconomic arbitrage opportunity is detected. A correction or convergence in
these relations to the long term equilibrium can realize a handsome profit. Besides
long/short positions in related instruments the exploitation of inefficiencies can also
lead to directional (long or short positions) bets. An example for such a trade is the
development of an asset or stock index relative to a macroeconomic variable. When
suddenly a divergence in the macroeconomic variable and the underlying price or
index value occurs, there is also an opportunity for a long/short arbitrage trade.260
As the mandates of macro hedge fund managers are very flexible we will not extract
separate risk factors for this group of strategies. The central success factor is the
254 See for example Ahl (2001) for a short introduction.255 See Lungarella (2004), pp. 10-11.256 See Burstein (1999), pp. 74-75.257 See Burstein (1999), p. 44.258 See Burstein (1999), p. 50.259 Not necessary on the same level but usually with a fixed difference.260 See further examples in Burstein (1999).261 See Burstein (1999), Ineichen (2000), pp. 42-44, and Lhabitant (2002), pp. 115-116.
3 Hedge Fund Strategies 72
strategy description
directional macro directional bets in different markets
anticipating macroeconomic developments
long/short macro bets on the spread of different instruments
anticipating macroeconomic developments
macroeconomic arbitrage long and/or short positions
based on macroeconomic variable(s)
Table 3.13: Macroeconomic arbitrage strategies261
ability of the manager to identify and make adequate use of trends or divergences.262
Besides the use of the right instruments, putting such strategies into action also
requires strong timing abilities of the manager.263 The use of leverage (directly or
through derivatives) is very common and usually extensive with macro strategies.
3.2.8 Short Selling
Short selling is an integral element of many hedge fund strategies.264 Most market
neutral strategies or at least partially hedged strategies make use of this opportu-
nity.265 But short selling is also constituting a self-contained investment discipline
in equities or whole equity indices when dedicated-short hedge funds focus to profit
from declining stock prices.
With short selling strategies a negative exposure to the market, an index and/or
the price of a single stock is set up. There are various possibilities to attain such
262 See Lhabitant (2002), p. 116, and Strome (1996), p. 29.263 See Strome (1996), p. 32.264 For examples on other users and uses of short selling see Financial Services Authority (2002b),
pp. 12-13.265 See for example the equity hedge strategies in Section 3.2.1 on page 35.
3 Hedge Fund Strategies 73
a negative exposure like selling borrowed securities, selling (index or single stock)
futures, buying put options or selling call options (see the trades in Table 3.14 on
the following page.).266 However the main instrument for hedge funds is the physical
short sale,267 consisting of the borrowing and sale of stock with the intention to buy
the shares back at some future date at a lower price.268 Short selling faces various
restrictions depending on the market place. In many emerging markets short sales
are completely prohibited269 and major US exchanges apply the uptick rule (Rule
10a-1 of the Securities Exchange Act) which prevents short sales unless the stock’s
last trade was at the same or higher price than the previous trade. Another point
is the tax discrimination of capital gains from short selling as these profits are fully
taxed.270
In an equity short sale transaction the resulting position has an equity and a fixed
income component and is therefore not the exact opposite of a long position in
the respective stock.271 This is a result of the technical short selling process. As
opposed to buying stocks, a short sale requires no investment or borrowed money
but has to be collateralized, usually with liquid government bonds like US treasury
securities. The revenue from the sold stocks is held in a restricted cash account
earning an interest rebate.272 The total profit or loss of such a short sale transaction
is influenced by four factors273: the isolated result of the short stock position, the
266 See D’Avolio (2002) for an excellent description and analysis of the US securities borrowing
(and lending) market.267 See Angel et al. (2003) for an interesting analysis of short selling on the NASDAQ. The authors
detected a higher frequency of short selling activity among stocks with high returns and among
actively traded stocks.268 The major tools for identifying short sale candidates are described in Section 3.2.1 on equity
hedge strategies. See for example Apfel et al. (2001) for a brief illustration of short sales.269 See Charoenrook/Daouk (2003) on short sale constraints and put options trading in different
developed and emerging equity markets.270 See for example Cottier (1998), p. 124. For detailed implications of the German tax legislation
on stock borrowing see Schmittmann/Schaffeler (2002) and Hahne/Obermann (2004).271 See Ineichen (2000), p. 45, and Ringoen (1996), p. 90.272 This is the so-called short rebate.
3 Hedge Fund Strategies 74
additional margin interest on this profit or loss,274 the dividend payments to the
security’s lender275 and the short rebate earnings.276
trade description
short selling physical short sales of overvalued stocks
futures selling selling stocks to a future date,
selling index or single stock futures
equity option trading long put options or put warrants,
short call options or call warrants
total return swaps synthetic short exposure to an index or stock
contracts for difference short position in CFD (synthetic swaps)
Table 3.14: Trades involving explicit or implicit short sales277
A negative exposure due to implicit or explicit short sales is set up with the trades
listed in Table 3.14. Total return swaps pay total return on some reference asset
with a life longer than the swap (for example a stock market index or single stocks).
This includes interest, dividends or fees and capital gains/losses in exchange for
floating interest payments. When a total return index (with reinvested dividends)
is considered, the contract can also be categorized as equity swap.278 Contracts
273 The interest income on the collateral is not considered, as this profit can not be attributed to
the short selling strategy. For an opposite view see Ringoen (1996).274 Depending on the development of the underlying stock, cash is released from the restricted
account, then earning a higher rate of return (declining stock price), or the restricted credit
has to be increased trough a margin loan at a higher interest rate (increasing stock price), see
Ringoen (1996), p. 92.275 See Schmittmann/Schaffeler (2002) for tax implications concerning dividend payments under
the German tax legislation.276 See Ringoen (1996), pp. 90-94.277 See Cottier (1998), pp. 123-124, Hull (2003), pp. 601-602, Ineichen (2000), pp. 44-45, Financial
Services Authority (2002b), pp. 9-14, and Mattinson (2002).278 See for example Hull (2003), pp. 601-602 and 644-645.
3 Hedge Fund Strategies 75
for difference (CFD), also known as synthetic swaps, are equity derivatives allowing
investors to participate in stock price, stock index or exchange traded funds (ETF)
movements without buying and selling the shares themselves. The value of the
contract is defined as a number of shares multiplied by the share price. Therefore,
the profit or loss in CFD is the difference between opening and closing values for the
contract. The central benefits of CFD are the exemption from the UK stamp duty
that is levied on standard equity trades (this is the origin of the financial contract)
and the relative ease of selling short.279
Share prices can fall to a value of zero but can (theoretically) reach an infinite
value. This unlimited downside potential in combination with the passive portfolio
strategy of backing losers and diminishing winners280 requires active management
of the short selling portfolio.281 Risk factors in short selling strategies with equity
instruments are summarized in Table 3.15 on the following page. The main impact
on the risk profile of short selling hedge funds has equity market risk. The exposure
to the equity market development is negative and therefore momentum is a large risk
factor to short sellers as overvalued stocks can continue to outperform.282 Credit risk
for the collateral is usually very low but the credit quality of the shorted company
influences the outcome of a short sale. The hedge fund is short default risk as the
trade will profit from the short position’s default.283 Therefore, an upgrade in the
company’s default risk will affect the short position in a negative way.
Liquidity risk plays an important role with short selling activities. When hedge funds
focus on large capitalization stocks, liquidity risk is less severe than with small and
279 See for example Mattinson (2002) and Connolly (2000).280 In a long only portfolio the passive strategy would overweight winning and underweight loosing
stocks.281 See Ringoen (1996), pp. 95-96.282 See Ringoen (1996), p. 94, and Ineichen (2000), p. 45. In the bull markets of the 1990’s
many short selling managers migrated to long/short or short biased equity hedge strategies
(Section 3.2.1 on page 35), see Lhabitant (2002), p. 120.283 See Ineichen (2000), p. 45.
3 Hedge Fund Strategies 76
mid caps as these stocks can be borrowed very efficiently.284 But also with these
intensely traded securities a short squeeze can occur for example in periods of rising
prices when short sellers even boost demand in order to cover their positions.285
The resulting high prices force short selling hedge funds to close their positions.
In general, liquidity increases when prices go up, so short sellers usually have to
cover their positions at higher prices in order to avoid (further) losses on their
positions.286 Another source of liquidity risk are active managers calling in the shares
they were lending to the short sellers. This is sometimes called an “engineered”
short squeeze.287 Therefore, re-calling the borrowed stock in a physical short selling
transaction is a particularly important risk factor for short selling funds as this might
force the manager to close positions.288 Interest rate risk is a minor influence. A
fluctuation of interest rates has some impact on the short selling as it could reduce
the short rebate on the restricted cash account.289
risk factor description/position
market risk - equities unlimited downside potential, momentum
market risk - interest rates short rebate
credit risk short position
liquidity risk risk of a short squeeze
event risk recall of the borrowed stocks
Table 3.15: Important risk factors in short selling290
284 See Ineichen (2000), p. 45.285 See for example Financial Services Authority (2002b), p. 10.286 See Ringoen (1996), p. 97.287 This enables active portfolio managers to fight back against short sellers that band together
against constituents of his portfolio, see for example Arulpragasam/Chanos (2000), p. 242.288 See Financial Services Authority (2002b), p. 10, and Ineichen (2000), p. 45.289 See Ineichen (2000), p. 45.290 See Arulpragasam/Chanos (2000), pp. 241-246, Ineichen (2000), p. 45, and Ringoen (1996), p.
97.
3 Hedge Fund Strategies 77
Security selection and the general market environment are the key factors influenc-
ing returns of this strategy.291 In addition short selling requires very efficient stock
borrowing facilities and timing skills.292 Diversification in short selling hedge funds
is achieved by positions in a significant number of stocks.293 Specific influences to
be diversified are for example exposures to certain industries, short-interest stocks,
companies near bankruptcy or stock split candidates.294 Leverage could be created
within the technical short selling process. As the securities in the collateral usually
account for at least the value of the stocks borrowed (due to the legal structure of the
borrowing), there is no real leverage involved in this transaction.295 When deriva-
tives are used to set up, the short positions leverage might be generated depending
on the margining regulations. In a portfolio context, short selling hedge funds are
very often used to balance long-biased managers in fund-of-funds structures.296
3.2.9 Emerging Markets
Emerging market hedge funds are grouped by a geographic relationship as they focus
on some specific geographic areas but not on a narrow band of trades. The term
“emerging” refers to economies as a whole, and in the hedge fund context it is based
on criteria like a medium gross national income (GNI) per capita297, future growth
291 See Ineichen (2000), p. 46, and Ringoen (1996), pp. 98-99.292 See Cottier (1998), p. 123.293 Ringoen (1996) speaks of 100 to 200 stocks in order to provide effective diversification, see
Ringoen (1996), p. 96.294 See Ringoen (1996), p. 97.295 See Financial Services Authority (2002b), p. 10.296 See Saerfvenblad (2002).297 The former measure for country classification used by the World Bank was gross domestic
product (GDP) per capita. Nowadays the world bank ranks economies in four GNI per capita
groups. The groups according to 2002 GNI per capita are: low income ($ 735 or less), lower
middle income ($ 736 - $ 2,935), upper middle income ($ 2,936 - $ 9,075) and high income,
(above $ 9,076), see World Bank (2004). The category upper middle income was previously
known as the emerging markets class.
3 Hedge Fund Strategies 78
potential and the existence of a public securities market with reliable data. Most
economies with such growth markets are located in Eastern Europe, Asia, Latin
America or the Middle East.298 The focus of hedge fund managers is on exploiting
inefficiencies in all types of securities. Reasons for these opportunities in emerging
markets are political and economic by nature.299 The low interest from international
investors is caused by political uncertainties and the low liquidity of many emerging
market securities.300
Hedge fund strategies in emerging markets have to compete with an increasing
number of traditional long-only fund managers looking for high returns and/or di-
versification. The Fund Rating Agency Lipper currently monitors over 200 emerging
market mutual funds, compared with about 80 emerging market mutual funds in
1995.301 So these markets are no longer an investment novelty, but the early 1990’s
expectations of growing liquidity, falling volatility and low correlations with devel-
oped markets were not completely fulfilled.302 In the first quarter of 2005 HFR
reportes a market share of emerging markets hedge funds of 3.5 percent.
Emerging market hedge funds take positions in emerging markets equity, corporate
or government debt and/or in corresponding derivative securities.303 Opposed to
macro hedge funds (see Section 3.2.7 on page 69), emerging markets managers do
not take on large macroeconomic bets. Emerging markets hedge funds buy and sell
financial instruments and try to hedge residual risk such as currency exposure or
market risk. The portfolios of most emerging markets hedge funds predominantly
consist of long positions. At least they exhibit a substantial long bias304 as most
298 See Lhabitant (2002), pp. 116-117.299 See Cottier (1998), p. 137.300 See Speidell (2002), p. 5.301 See Speidell (2002), p. 9.302 See Speidell (2002), pp. 5 and 13.303 See for example Cottier (1998), p. 137.304 See Ineichen (2003), p. 349.
3 Hedge Fund Strategies 79
of the emerging markets allow for only limited short selling and do not offer liquid
futures contracts or other derivative instruments.305 So, hedge funds often have to
move to OTC trades (like for example equity swaps) in order to hedge risk factors
or set up negative exposures.306
Emerging market funds employ a mixture of fixed income trades, convertible arbi-
trage and equity hedge trades with a regional focus. Therefore, the trades in this
category very often overlap with the strategies mentioned before in sections on eq-
uity hedge (Section 3.2.1), convertible arbitrage (Section 3.2.2) and fixed income
arbitrage strategies (Section 3.2.3). Some characteristic trades of emerging market
hedge funds are summarized in Table 3.16.
trade/strategy description
long/short countries long and/or short positions in different countries
equity trades equity hedge strategies or levered long positions
convertible trading long undervalued convertible bonds
debt trades broad spectrum of fixed income trades,
high yield trading, credit arbitrage
Table 3.16: Emerging markets trades307
It is obvious that emerging market hedge funds can be exposed to the full range of
risk factors as the managers are not restricted to a certain asset class. Numerous
risk factors in emerging market trading relate to the market environment. Emerg-
ing markets depend on capital inflows in order to finance their economic growth. If
investors get more risk averse and withdraw their capital from emerging markets or
stop depositing funds, this might lead to severe problems in these economies and
305 See Ineichen (2000), p. 47, and Charoenrook/Daouk (2003) on short sale constraints and put
options trading in emerging equity markets.306 See Lhabitant (2002), p. 117.307 See for example Cottier (1998), p. 138, Ineichen (2000), p. 47-48, and Speidell (2002).
3 Hedge Fund Strategies 80
the corresponding capital markets.308 Due to the common long positions in equi-
ties, hedge funds are often exposed to market risk. As the dependence structure in
emerging markets is usually less stable than in developed economies, changes in cor-
relation are another important (market) risk factor. Furthermore, the correlations
among different stocks are significantly higher than in developed markets. This leads
to smaller diversification benefits in emerging market equity portfolios.309 When the
hedge fund trades debt instruments, the results are also sensitive to changes in inter-
est rates. But the main risk factor in emerging markets fixed income trades is credit
risk. Investors are usually long the issuer’s and the emerging economy’s default
risk.310 As these emerging markets are usually less liquid than their counterparts in
developed economies, liquidity risk plays a major role with emerging market hedge
funds, too. Emerging market funds have long (biased) positions in inefficient emerg-
ing markets through the traded (less liquid or even illiquid) securities.311 They
provide and enhance liquidity and are therefore subject to liquidity risk.312 Fur-
thermore, political and legal uncertainties in emerging economies are sources of risk
as subcategories of event risk. This event risk is high in many emerging markets
hedge funds.313 The most important risk factors in emerging markets hedge funds
are listed in Table 3.17 on the next page.
308 See Signer (2003), p. 31.309 See Ineichen (2003), p. 349.310 See Ineichen (2000), p. 47.311 A further threat to the liquidity of emerging markets is the global trend to float weighted
benchmarks as this will have a significant effect on capital allocation due to the small free
float in emerging markets, see Speidell (2002), p. 5. For the market capitalization of (Asian)
emerging markets in the MSCI World free float index see for example Gruenig (2002), p. 10.312 See Ineichen (2003), p. 349.313 See Cottier (1998), p. 137.314 See Cottier (1998), p. 137 , Ineichen (2000), p. 47, and Speidell (2002).
3 Hedge Fund Strategies 81
risk factor description/position
market risk - equities long or at least long biased positions,
increasing volatility, correlation jumps
market risk - interest rates trading emerging markets debt
credit risk long the issuer’s and the country’s default risk
liquidity risk long market and mispriced securities
event risk political and legal risk
Table 3.17: Important risk factors in emerging markets hedge funds314
3.2.10 Managed Futures
Managed futures funds take long and short positions in (liquid) futures and options
contracts on financial instruments (currencies, interest rates, stock market indices)
and various physical commodities. Some managed futures funds additionally trade
OTC derivatives.315 Managed futures investments are sometimes referred to as
derivatives funds, futures funds, managed derivatives or leveraged funds,316 and they
are seen as a distinct asset class by some authors.317 The money managers are in
most cases commodity trading advisors (CTA) or commodity pool operators (CPO)
that are registered with the US Commodity Futures Trading Commission (CFTC)
through membership in the National Futures Association or registered with the
Securities and Futures Authority in the UK.318 Further managed futures vehicles are,
for example, the so-called public commodity funds.319 CTAs manage single managed
futures funds while CPOs coordinate commodity pools that are organized as private
partnerships by selecting professional traders and/or CTAs.320 The registration of
315 See Jaeger (2002), p. 96.316 See Cottier (1998), p. 10.317 See for example Chance (2000), pp. 14-21.318 See Hull (2003), p. 34, Harding/Nakou (2004), p. 64, and Chance (2000), pp. 49-50.319 See Edwards/Liew (1999), p. 46.
3 Hedge Fund Strategies 82
managed futures is a major difference to hedge funds as these are largely exempt
from government regulations.
There is a natural link between several hedge fund strategies and managed futures
trading. The link to emerging market hedge funds (see Section 3.2.9 on page 77)
is established as currencies of the emerging markets segment are frequently traded
in managed futures funds.321 Furthermore there are intersections with global macro
strategies (see Section 3.2.7 on page 69) because managed futures trades also analyze
macroeconomic information in order to anticipate the moves of certain markets or
securities.322 Currency trading funds are also very similar to managed futures or
global macro strategies. Both types of funds trade in foreign exchange markets
employing currency spot and derivatives products.323
The universe of managed futures strategies can be divided in three broad categories
(see Table 3.18 on the following page). Systematic trend-following managers
try to identify trends in prices using quantitative models.324 According to Lungarella
(2002) and Lhabitant (2002) about 60 to 70 percent of managed futures funds follow
such a systematic trend-based approach.325 They are usually active in very liquid
markets with low transaction costs. The individual trades depend on the systematic
application of models and the resulting signals to buy or sell. These decisions are
often based on trendfollowing models employing momentum indicators, but coun-
tertrend and spread models are also widely employed.326 The computerized trading
systems often work with protective stops that are usually adjusted on a daily basis
in order to close losing positions quickly.327 Although most trades from such an
320 See Edwards/Liew (1999), p. 46.321 See Yang/Faux (1996), p. 87.322 See Jaeger (2002), p. 96, or Yang/Faux (1996), p. 81.323 See Cottier (1998), p. 141, and Jaeger (2002), p. 96.324 For an in-depth quantitative analysis of trend-following strategies see Fung/Hsieh (2001).325 See Lungarella (2002), p. 11, and Lhabitant (2002), p. 120.326 See for example Lungarella (2002), p. 10, or Jaeger (2002), p. 97.327 See Lungarella (2002), p. 11.
3 Hedge Fund Strategies 83
automated system are unprofitable, this stop loss strategy is able to produce attrac-
tive returns as the managers is only keeping at profitable trades.328 Discretionary
managers or funds make use of fundamental analysis and/or superior information.
Very often these funds focus on a market or sector where the manager has exten-
sive experience. A discretionary trader has to be familiar with the factors that
can affect prices in the particular market. In addition to knowledge about such
fundamentals for this market and the changing relationship of demand and supply,
managed futures traders often make use of technical trend-models to determine the
optimal timing for entry or exit.329 A third passive futures strategy is introduced
in Jaeger et al. (2000). The strategy tries to capture the premium commercial
hedgers are willing to pay for insurance on different markets. When the manager
identifies trends, he offers supply in hedging instruments matching the demand of
potential investors. Therefore, these passive strategies usually depend on long-term
trendfollowing models.330
strategy description
systematic trend-following technical trend-, countertrend-
or spread-based trading
discretionary strategies fundamental (and technical) approaches
in special segments
passive futures strategies capturing hedging premia
Table 3.18: Managed futures strategies331
As the trades of systematic trend-following and discretionary traders are ranging
from short term trend- or countertrend-following over directional volatility trading
328 See Jaeger (2002), p. 97.329 See Jaeger (2002), p. 100, and Lungarella (2002), p. 11.330 See Jaeger et al. (2000) and Jaeger (2002), p. 102-106.331 See Lungarella (2002), p. 10, Jaeger (2002), pp. 100-106, and Cottier (1998), pp. 141-146.
3 Hedge Fund Strategies 84
to currency arbitrage, these trades are subject to a very diverse set of risk factors.332
Market risk moves are the dominant risk factor with most managed futures trades
since they usually have a significant exposure to different markets. But especially in
non-trending market environments with frequent price reversals such trend-following
strategies are problematic because the implemented models need persistent price
trends to be profitable.333 In general, managed futures tend to perform well under
extreme (trending) market conditions.334 Liquidity risk is also an important issue
with managed futures as the funds provide liquidity to different derivatives markets.
This risk premium is an important part of the fund’s return for long-term futures
strategies.335 For systematic trend-following managers, model risk is an important
component of the funds risk profile. As the structure and dependencies of financial
markets continuously change the quantitative models that have been fitted (and
probably over-fitted) to a historic data set and market environment must be adjusted
on a regular basis.336 Finally, event risk has to be considered with managed futures.
Especially with discretionary trading strategies, the impact of external economic or
political decisions on the fund’s returns can be devastating. These risk factors are
aggregated in Table 3.19 on the next page.
3.2.11 Closed-end Fund Arbitrage
Closed-end fund arbitrageurs try to capitalize on the difference between the net
asset value and the trading price of closed-end fund shares.338 A closed-end fund
332 See Jaeger (2002), pp. 97 and 100, for examples of trend-following and discretionary trades.333 See Jaeger (2002), pp. 98-99.334 See Harding/Nakou (2004), p. 64.335 See Jaeger (2002), p. 98, and Harding/Nakou (2004), p. 63.336 See Jaeger (2002), pp. 98-99.337 See Jaeger (2002), pp. 100-106.
3 Hedge Fund Strategies 85
risk factor description/position
market risk long or short positions in commodities
or financial instruments
liquidity risk managed futures are liquidity providers
model risk flawed or unsuited trading models
event risk economical and political events
Table 3.19: Important risk factors in managed futures337
is one of the basic types of investment companies and is registered under the US
Investment Company Act of 1940. In Section 5 of the Investment Company Act,
a closed-end company is defined as management company other than an open-end
company. The fund’s capital is mainly raised through an initial public offering with
a fixed number of shares. The limited number of outstanding shares is the main
difference to open-end funds where the quantity of shares constantly changes as they
can be issued or redeemed at the fund’s particular net asset value. The number of
closed-end fund shares can in fact change for some specific reasons, for example
if the fund offers rights to purchase additional shares, if the fund buys-back and
retires shares or if the fund issues shares in a dividend reinvestment programm.339
If the fund trades at a premium, there is also the possibility of secondary stock
offerings.340 Therefore, closed-end fund shares have to be traded on the secondary
market, where no instantaneous arbitrage mechanism is in place in order to adjust
338 See Lhabitant (2002), p. 95. Net asset value arbitrage in open-end funds is often connected
with illegal or at least unwanted trading practices like late trading (prohibited by the Martin
Act and SEC regulations (rule 22c-1)) or timing strategies and is therefore not subject to this
section. For an insightful investigation concerning the Canary hedge fund see Supreme Court
of The State of New York (2003) and for the consequences of timing activities in mutual funds
see for example Zitzewitz (2003). Zitzewitz (2003) estimates a dilution to long-term mutual
fund shareholders of over $4 billion per year caused by timing arbitrage.339 See for example Bush (2003), p. 1, and Maier/Brown (2000), p. 5.340 See Fredman/DeStaebler (2000).
3 Hedge Fund Strategies 86
the fund’s price to its intrinsic value. So, market forces determinate the closed-end
share price, which may be greater or less than the shares’ net asset value, depending
on the relation between demand and supply.
Closed-end funds invest in equity, convertible or fixed income securities with the
whole spectrum of different investment philosophies.341 In 2002, the closed-end
fund industry included over 500 funds with approximately $ 150 billion in assets.342
Despite of some advantages of closed-end over open-end funds,343 a large part of
these closed-end funds trade at a discount to their net asset value.
The core of an arbitrage strategy with closed-end funds is about identifying closed-
end funds trading at substantial discounts to their net asset value, whose underlying
portfolio can be easily hedged in order to attain market neutrality. Abnormal dis-
counts (and premiums) represent market inefficiencies. The identification of real
mispricings is based on a thorough analysis of the fund, examining the fund’s hold-
ings, dividend yield, performance and management, expenses, number of sharehold-
ers and its historical discount (or premium).344 Trading strategies with closed-end
funds usually result in long positions of closed-end funds under, at least partial,
hedging of the fund’s assets. To a very large extent, such strategies focus on timing
as buying closed-end funds is especially promising in some situations. Deep dis-
341 The types of closed-end funds enclose municipal bond funds (54 percent), taxable bond funds (17
percent), international equity funds (12 percent), international bonds funds (8 percent) and US
equity funds (9 percent), see Closed-End Fund Association (2002), p. 12 (data: invested capital
according to Wiesenberger, a Thompson Financial Company, July 2002). See Fredman/Scott
(1991) for a review of different fund types.342 See Gabelli (2002).343 Advantages of closed-end funds are for example lower marketing and distribution expenses,
the possibility of using leverage and a stable pool of capital with lower cash flow risk allowing
for long term investments and hight total return. Disadvantages can be seen in the premium
at the shares public offering, the lack of liquidity in some funds’ secondary market or in the
increased risk due to leverage, see Fredman/Scott (1991), Bush (2003), Maier/Brown (2000)
and Closed-End Fund Association (2002).344 See for example Fredman/Scott (1991), pp. 387-390.
3 Hedge Fund Strategies 87
counts are offered in generally oversold markets or at the end of the year when other
investors realize their potential tax losses in funds. Special situations in a fund’s life
could eliminate the common discount in trading prices. Such developments are for
example a share repurchase, the replacement of managers, the (forced) conversion to
an open-end fund or the liquidation of the whole fund.345 Even if no immediate or
complete correction in the closed-end fund’s price occurs, the income dividends or
capital gains distribution with these undervalued closed-end funds is at least higher
in relative terms (as they are paid on the assets as a whole).346
With the historical observation that most funds do not change between discount
and premium, other long/short strategies with closed-end funds have been devel-
oped. Long positions in high discount funds while shorting high premium funds is
betting on (slight) reversals in the premium or discount of closed-end funds. Under
the assumption that the market offers a premium for good and a discount for bad
performing funds, buying closed-end funds with a low premium and shorting funds
with a low discount is an appropriate strategy.347
risk factor description/position
market risk - equities usually hedged
market risk - interest rates usually hedged
liquidity risk small issues show low liquidity
event risk management opposition,
anti-takeover provisions
Table 3.20: Important risk factors in closed-end fund arbitrage348
345 See Fredman/Scott (1991), pp. 381-387.346 See Fredman (2000) and Closed-End Fund Association (2002).347 See Dauenhauer et al. (2000) for an empirical study on trading strategies.348 See Fredman/Scott (1991), pp. 381-397, Closed-End Fund Association (2002), p. 11, and
Maier/Brown (2000), p. 7.
3 Hedge Fund Strategies 88
Risk factors in closed-end fund arbitrage are summarized in Table 3.20 on the pre-
ceding page.349 Market risk emerges with equity and fixed income closed-end funds
but hedge fund managers attempt to hedge both risk components. The volatility of
closed-end prices is usually higher than the volatility of comparable mutual (open-
end) funds as they will almost never trade at their net asset value. In addition, the
leverage amplifies the volatility. As closed-end funds are exchange-traded, liquidity
is not a major risk factor for most large funds, but some small sized issues do not
trade significantly imposing a greater extent of liquidity risk.350 Legal risk with
closed-end trades references to the opposition of the closed-end fund managers in
general and the specific anti-takeover provisions for most funds. Funds that are rel-
atively easy to convert to open-end funds usually trade at narrower discounts than
their harder to open pendants with a larger amount of this legal risk.351
3.2.12 Depository Receipts Arbitrage
Arbitrage with depository receipts (DRs) comes close to the real meaning of (risk-
less) arbitrage, the simultaneous buying and selling of a security at two different
prices in two different markets. As depository receipts arbitrage involves positions
in equity instruments, it could also be classified among equity hedge strategies (see
Section 3.2.1 on page 35).
A DR is a physical certificate evidencing ownership in a specified number of de-
pository shares from a company outside the market in which the DR is traded.
Depository shares represent several underlying shares that are deposited with a cus-
todian bank in the issuer’s home market. The main advantages of depository receipts
are that there is no currency conversion in trading and in receiving dividends and
349 See Closed-End Fund Association (2002) or Fredman/Scott (1991) for further details on invest-
ment risk with closed-end funds.350 See Maier/Brown (2000), p. 7, and Bush (2003), p. 2.351 See Fredman/Scott (1991), pp. 390-391.
3 Hedge Fund Strategies 89
they help in minimizing higher overseas transaction costs and custodial fees. They
are frequently identified by the markets in which they are available. DR that are
primarily available to US investors are called American depository receipts (ADRs),
privately placed DR are named Rule 144A ADRs352 and global depositary receipts
(GDRs) are offered in two or more markets outside the issuer’s home country.353
GDRs are usually listed on the London or the Luxembourg stock exchange, while
ADRs are listed on US exchanges. For investors, DRs provide an alternative to
investing in foreign equities directly without the inconveniences such as currency
conversion and foreign settlement procedures.354
Taking into account the current exchange rate, these depository receipts’ trading
price sometimes differs from the value of the underlying shares in the company’s
home or some other market.355 Explanations for gaps in the trading prices are trans-
action cost and non-synchronous trading times between the different markets.356 In
these situations a hedge fund’s manager will bet on the convergence of the prices
with long and short positions in these securities.357 The two-way fungibility (DRs
can be converted into underlying shares and foreign shares can be re-converted into
DRs)358 offers another direct arbitrage possibility. An investor, who compares the
DR price with the price of the underlying share, can lock in a profit if the price
differential (in both directions) is sufficiently large to cover transactions costs.359
352 Rule 144A of the Securities Act of 1933 regulates private resales of securities.353 See Citibank (1999), p. 2, or Mishkin/Eakins (2006), p. 290.354 See for example Alaganar/Bhar (2001), p. 98.355 Even though the depository receipts are dollar denominated they do not eliminate currency
risk.356 See Kim et al. (2000), pp. 1362-1363.357 See Lhabitant (2002), p. 97.358 See for example Citibank (1999), pp. 4-7.359 For a detailed analysis of tax effects and tax-induced trading activity with ADRs see
Callaghan/Barry (2003).
3 Hedge Fund Strategies 90
3.2.13 Regulation D
Hedge fund managers following Regulation D-360 or so-called PIPE-361 strategies
make private placements in public companies in need of financing.362 These private
equity investments that are placed with small capitalization companies, are very
often directly negotiated and usually take the form of equity issues, debenture issues,
convertible issues or the issue of warrants in return for the capital allocation.363 In a
direct equity issue, the hedge fund will purchase the company’s stock at a discount
with respect to the market price or to an anticipated market price. Convertibles can
be issued with a floating strike price or subject to a look-back provision, making
the private placement market neutral.364 The convertible securities or debentures
with the right to be exchanged into common stock are usually issued at a discount
or bear high coupons. Hedge fund trades involving Regulation D securities consist
of long positions in the private placements. In the case of fixed exercise prices for
convertibles, the stock’s market risk is often hedged with short positions in the
underlying shares.
Regulation D trades are profitable because of the discount in the issued securities.
This discount is offered in exchange for the risk factors an investor has to bear.
The key risk factor with Regulation D debenture trades is default risk, if the issuer
is unable to meet it’s principal or coupon obligations.365 During a limited period
of time after the securities’ issuance366 they are only tradeable among accredited
investors until the registration statements are filed with the SEC. The liquidity risk
360 The Regulation D is part of the Securities Act of 1933 since 1982 and offers an exemption from
registration requirements for private placements.361 PIPE stands for Private Investments in Public Equities.362 Most US hedge funds make use of Regulation D themselves when placing their securities directly
to individual investors, see Lhabitant (2002), p. 119.363 See for example Jaeger (2002), p. 73.364 See Jaeger (2002), pp. 73-74.365 See Jaeger (2002), p. 77.366 This period ranges from 75 to 180 days, see Jaeger (2002), p. 74, and Lhabitant (2002), p. 120.
3 Hedge Fund Strategies 91
with these securities is especially high in this pre-registration period when there
is no market to liquidate the position.367 As the issue’s capitalization very often
represents a significant part of the firm’s overall market capitalization, the securities
are also subject to post-registration liquidity risk as a consequence of the market
impact.368 Equity market risk is a relevant risk factor for convertibles with a fixed
exercise price, but if the underlying stocks are available, this market risk is usually
hedged.
3.3 Fund of Hedge Funds
To screen the hedge fund universe, to select and monitor managers is a difficult and
quite time consuming task for investors. The evaluation process is very complex
and usually a lot of experience is necessary to analyze the organizational structure
of hedge funds and the implemented trading strategies. At this point funds of hedge
funds can be an interesting investment opportunity as they provide an alternative
access to hedge funds.369 A fund of hedge funds manager blends funds from the
same strategy group or funds that pursue different strategies in order to diversify
over risk factors and unsystematic risk. Fund of hedge funds therefore allow investors
to obtain an instant exposure to a diversified hedge fund portfolio.
According to hedge fund databases, around 25 percent of the reporting funds were
classified as fund of hedge funds.370,371 The question is whether fund of hedge funds
managers are really able to add value for the investor. Finding the right or best fund
367 See Jaeger (2002), p. 77.368 See Jaeger (2002), p. 77.369 See UBS (2004), pp. 17-18, for recent figures on the fund of hedge funds industry.370 As of the first quarter of 2005 according to HFR.371 It is interesting to note that the fund of hedge fund universe is very concentrated. The 20
largest funds of hedge funds have a market share of about 80%, see Mercer Oliver Wyman
(2005), p. 17.
3 Hedge Fund Strategies 92
of funds manager is still a challenging task, but the fund of hedge funds universe
is much smaller than the universe of hedge funds.372 At first we will examine some
advantages and disadvantages of fund of hedge funds. The major benefits and
drawbacks of fund of hedge funds are summarized in Figure 3.3. Then we will take
a closer look at the investment process for hedge funds that is important for the
understanding of the fund of hedge funds management and the selection of hedge
funds in general.
����������������
��������� ����������
� ���������������������
����� �������
����������������������
��������������
������������������
���������������������� ���
������������������������
������������� �!����������
�������������������
"����� ��������
Figure 3.3: Advantages and disadvantages of funds of hedge funds
3.3.1 Advantages of Funds of Hedge Funds
Several advantages of fund of hedge funds can be identified. Some of these advan-
tages are close to the arguments for rather investing in mutual funds than in single
stocks. But we will also introduce some benefits of fund of hedge funds that deal
with the characteristics of the hedge fund industry.373
372 See for example Stemme/Slattery (2002), p. 66-68, for the investment process with funds of
hedge funds.
3 Hedge Fund Strategies 93
Investors try to diversify their hedge fund exposure and allocate money to several
funds and different hedge fund strategies. An argument for funds of hedge funds
is that they are able to provide efficient allocation and diversification in a
large number of hedge funds.374 The selection of hedge funds by fund of funds
managers is usually much more efficient as fund of hedge funds constantly screen
the large hedge fund universe and provide standardized due diligence procedures.375
The same holds true for the ongoing monitoring process with target hedge funds.376
Additionally, fund of hedge funds managers provide investors with their expertise
in optimizing hedge fund allocations.377 Due to their experience, most fund of
hedge funds managers should be able to estimate return distributions and return
dependencies more accurately than the individual investor.378 Besides the selection,
optimization and monitoring of hedge fund allocation, fund of hedge funds managers
also simplify the process of allocating money to target funds. In the process of
changing the exposure to a diversified portfolio of hedge funds, a fund of hedge
funds investor does not have to deal with the different investment or subscription
policies of every single target fund. The same holds true for the disinvestment
process. In contrast to an investment in the same number of target funds, a fund of
hedge funds investor faces only one redemption clause.379,380
Risk control and return predictability are important reasons for fund of hedge
funds investments. Active management of the fund of funds should provide risk con-
373 See for example Section 2.3 for important hedge fund industry characteristics.374 See for example Brown et al. (2004), Table 1, or the simulation approach in Wintner (2001)
for the diversification effect with hedge funds. An empirical analysis with regard to higher
moments of return distributions can be found in Kat (2002b).375 See Fothergill/Coke (2001), p. 13.376 See Ineichen (2000), pp. 95-96, for estimates of costs for hedge fund research.377 See for example the multiple objective approach in Davies et al. (2004) to determine optimal
allocations for a fund of hedge funds.378 See Lhabitant (2002), pp. 199-200.379 See Lhabitant (2002), pp. 196-199.380 See Section 2.3.6 on the liquidity of hedge fund investments.
3 Hedge Fund Strategies 94
trol381 and ensure capital preservation to conform with the absolut return approach
of hedge fund investing (see Section 2.3.3). In addition, the results of fund of hedge
funds are usually more stable over time and therefore better to predict than the
returns of single funds.382
Capacity constraints are common in the hedge fund industry (see Section 2.3.6
on page 23). Especially successful hedge fund managers close their funds to new
investors when they reach a certain size.383 Here fund of hedge funds could again
offer benefits to investors as they can provide access to closed funds. If the
fund of hedge funds is already invested in a closed fund, an allocation in the fund
of hedge funds implies an exposure to the desired hedge fund.384 In the case of a
so-called “soft close”, single hedge funds are officially closed to new investors, but
still accept capital from high-quality investors with a long-term focus and a sizeable
commitment.385 Here a fund of hedge funds manager might be able to allocate
capital while the requested amount is too large for an individual investor.386
In addition, funds of hedge funds provide diversification at a smaller minimum
investment.387 The minimum investment for fund of funds is usually only a fraction
of the minimum investments for single hedge funds.388 For an investor who composes
a portfolio with sufficient diversification, the minimum investment is multiplied and
381 See for example Jaeger (2000) for a general discussion of risk measurement in fund of hedge
funds.382 See Ineichen (2000), p. 94.383 See Section 3.1.1 on page 29 for more details on size risk.384 An important consequence is a dilution effect for existing investors, as the fund of hedge funds
manager is unable to allocate the capital inflow in the desired hedge fund in order to keep it’s
weighting constant. Therefore, fund of hedge funds are often closed when their core underlying
funds are closed, see for example Lhabitant (2002), p. 200.385 See Ineichen (2003), p. 409, and Brown et al. (2004), p. 1.386 See Lhabitant (2002), p. 200.387 Investors can not take this advantage of funds of hedge funds in so-called “private fund
of hedge funds” products that are customized for the needs of the individual investor, see
Hennessee/Gradante (2002).388 See Fothergill/Coke (2001), p. 12.
3 Hedge Fund Strategies 95
therefore even less attractive.389
Accounting and standardized reporting are further operational aspects that
advantage fund of hedge funds.390 The fund of hedge funds investor does not have
to collect and aggregate data from the target funds which is quite cumbersome
and time consuming. Usually there are systems in place to ensure a standardized
reporting from target funds to the fund of hedge funds manager. This allows funds
of hedge funds to report on different levels of aggregation (for example by position,
strategy or risk factor allocation).391 The information that is needed by investors
can therefore be delivered in time. Standardized (monthly) reports usually include
figures and ratios on the performance development for the fund of hedge funds and
the target fund level (net asset values and risk/return figures), a description of the
current situation and an outlook for the funds and the market environment.392 Some
sophisticated funds of hedge funds managers even offer web-based access to allow
their investors to monitor the portfolio construction.393
3.3.2 Disadvantages of Funds of Hedge Funds
Besides the advantages of funds of hedge funds that have been outlined in Section
3.3.1, there are also a few drawbacks of a fund of hedge funds structure that investors
have to consider.
The double fees structure is often seen as a major disadvantage of fund of funds
structures.394 The service and the advantages of a fund of hedge funds management
389 An investor that is diversifying across 20 single hedge funds with a minimum investment of
$500,000 each, would need at least an amount of $10,000,000 for a diversified hedge fund
allocation.390 See Mahadevan/Schwartz (2002), p. 49.391 See GlobeOp (2003), p. 70.392 See Schiendl/Hofer (2003), pp. 40-41. For recommendations on the reporting of funds of hedge
funds, see Amenc et al. (2004), pp. 12-26.393 See Wilmot-Smith (2003), p. 66.
3 Hedge Fund Strategies 96
comes at a certain price. A fund of hedge funds imposes additional management
fees and usually also performance fees. These fees are therefore charged twice,
on the target fund and on the fund of funds level.395,396 The degree of technical
knowledge that is required to properly understand and evaluate individual hedge
funds is sizeable.397 Therefore, this second level of fees can be compared to cost
savings in order to determine if a value is added by fund of hedge funds managers.398
Empirical research in this field determines a significant underperformance (absolute
and risk-adjusted) of funds of hedge funds compared to portfolios of single hedge
funds that can not be explained with research costs or biases.399
Lack of control over the actual allocation is another problem with fund of hedge
funds structures. The exposures to target funds are solely determined by the man-
agement of the fund of hedge funds. An individual investor is (usually) unable to
influence the strategy or target fund allocations. Therefore, the fund of hedge funds
manager could for example miss an interesting fund in the search process, screen
too restrictively or overlook qualitative risk factors in target funds.400 Selling or
redeeming the stake in the fund of hedge funds would be the only way to change
the investors risk profile. Moreover the fund of hedge funds management itself has
only limited control over the trades of target funds. Therefore, problems like the
cancellation of trades in the fund of hedge funds portfolio (because of long and short
positions in different target funds) or the duplication of trades in target funds might
394 See for example Lhabitant (2002), p. 202.395 See for example Barra RogersCasey (2001), p. 12.396 There are also some other streams of income for a fund of hedge funds: If the target fund shares
management or performance fees with the fund of funds (retrocession agreement), if the fund of
funds forces the target fund to use its clearing broker usually a commission is paid (kickback),
if a target fund offers a trailing fee when a fund of funds remains invested after the lock-up
period. See Lhabitant (2002), p. 202, and Fothergill/Coke (2001), p. 13.397 See Moore/O’Brien (2002), p. 30.398 See Ineichen (2000), pp. 96-97.399 See for example Brown et al. (2004), pp. 8-13, and Liang (2003).400 See Mahadevan/Schwartz (2002), p. 49.
3 Hedge Fund Strategies 97
come up.401
The dependence on other investors is a disadvantage especially for allocations
in smaller funds of hedge funds. Because of the advantageous redemption policies
(in comparison to the target funds, see Section 3.3.1), the fund of hedge funds
management has to maintain a higher level of liquidity.402 This allocation (money
market account) does not offer the same return to the investor as is expected from
the hedge fund portfolio. Even worse consequences for investors might arise if the
fund is forced to liquidate target fund positions to meet redemptions as this destroys
performance.403 The problem of this dependence on other investors could be avoided
if the fund of hedge funds gets listed on a secondary market.404,405
3.3.3 Hedge Fund Selection
As already mentioned, selecting hedge funds is a challenging task. For fund of hedge
funds as well as for direct hedge fund investors, a structured investment approach
is essential. This section describes the investment process for hedge funds that is
usually implemented in fund of hedge funds structure to avoid at least some of
the industry specific risks outlined in Section 3.1.1.406 Figure 3.4 on the following
page introduces an idealistic investment process which includes the most important
components concerning the investment process and the selection criteria for hedge
funds.
The hedge fund universe sets the frame for searching hedge fund managers. The
universe has grown substantially over the last years and is estimated to contain
401 See Lhabitant (2002), p. 205.402 See Lhabitant (2002), pp. 203-204.403 See Spitz (2001), p. 61.404 See Fothergill/Coke (2001), p. 12.405 See for example ABN AMRO (2004), for an overview on exchange-listed funds of hedge funds.406 See Seides (2002) for an interesting discussion of risk transparency in hedge funds.
3 Hedge Fund Strategies 98
about 6100 single hedge funds.407 The problem is to identify possible target fund
managers as they are not allowed to promote their funds in public. Nevertheless
there are many sources of information on hedge funds.
��������������� �
������������������ �
�����������������������
�������������������
���������������� � ���� ���� ���� ��������������� �
�����������������
���������������
���������������
��������������� ������������������������������������������ ����! ��� "#�� ������
Figure 3.4: Typical hedge fund selection process408
Publicly available hedge fund databases (see Section 4.1 on hedge fund databases)
are a starting point for the hedge fund search. But none of these databases can offer
the complete universe of hedge funds as the registration and reporting happens on
a voluntary basis. Additionally, many top hedge funds do not report to any of the
commercial data providers.409 Therefore, investors, fund of hedge funds managers
407 See HFR (2005), Barra RogersCasey (2001), p. 1, or Wilmot-Smith (2003), p. 65.408 See for example Lhabitant (2002), p. 186, for a related figure.409 See Lhabitant (2002), p. 187.
3 Hedge Fund Strategies 99
and consultants usually extend databases from several information providers by ac-
tively searching for new hedge funds. Manager and fund names are gleaned from
different sources like for example industry contacts, industry journals, recommenda-
tions, conferences and seminars or even cold calls from hedge fund managers.410,411
Such an ongoing screening process is usually followed by a pre-selection of funds
according to some quantitative and qualitative minimum requirements.412 Fil-
tering according to criteria like fund strategy (especially when a fund of hedge funds
is limited to certain strategies), minimum assets under management, fund domicile
and management experience leads to the pool of hedge funds that a potential
investor examines in greater detail.413
Before individual funds are selected from such a reduced database, the investor has
to determine his views on the economic development in general, on the market
environment or even on individual risk premia. This is a very important part of
the hedge fund selection process as it predetermines the strategy allocation. The
second step of such a top-down approach is therefore to analyze the attractiveness
of certain strategies or strategy groups in the anticipated market environment or
in stress scenarios.414 Here favorable and unfavorable market conditions and their
consequences on returns and the dependence structure of different strategies are
examined. Like the screening of the hedge fund universe, this is an ongoing process,
and major changes in the market environment should directly influence the portfolio
allocation.415
Based on the investor’s assessment of different markets, further quantitative and
410 See Jaeger (2002), p. 241.411 Some skilled investment managers are even identified and tracked before they actually launch
a hedge fund, see Ineichen (2003), p. 440.412 See for example Perisse (2004), p. 17.413 See Lhabitant (2002), p. 185.414 See Jaeger (2002), p. 236.415 See Jaeger (2002), p. 237.
3 Hedge Fund Strategies 100
qualitative analysis is carried out in order to condense the manager pool and arrive
at the so-called hedge funds short list.
Quantitative hedge fund analysis relies on historical time series of net asset val-
ues. It is therefore usually standardized and not too costly.416 In order to reveal
significant information behind the time series, the statistical evaluation of net asset
values needs a certain length of a fund managers track record. For very short histo-
ries of net asset values or returns this quantitative part of the analysis should not
be overly weighted. The time series of returns are usually analyzed on a stand alone
basis in order to obtain descriptive statistics and risk measures like volatility or
maximum drawdown.417 But more important is the analysis of hedge fund returns
in the context of market indices, risk factors and the returns of strategy benchmarks.
The fund’s returns in different market situations might reveal interesting results es-
pecially when the desired diversification in a broader portfolio context with bonds
and stocks tends to break down in extreme market situations.418 The analysis of
risk factors and their explanatory power for fund returns is another important point.
In a first step the size and the stability of the fund’s exposures to risk factors can
be analyzed over time and used for a peer group comparison.419 The second step
of such a risk factor analysis is to detect the manager’s ability to produce excess
returns after the adjustment for risk premia. This analysis of risk premia and the
corresponding “alpha” is very important to identify really skilled managers.
As quantitative analysis is a backward-looking process based on historical perfor-
mance, investment decisions should also consider qualitative aspects. These qual-
itative aspects are a pivotal part of the due diligence process. Sources of informa-
416 See Lhabitant (2002), p. 188.417 See Jaeger (2002), p. 238.418 See for example Ang/Chen (2002), Campell et al. (2002), Konberg/Lindberg (2001), and
Longin/Solnik (2001), on conditional correlations. The problems with correlations are also
addressed in Section 5.2.3 on page 184.419 See Lhabitant (2002), p. 188.
3 Hedge Fund Strategies 101
tion are primarily the offering memorandum of a fund (stating information about
the fund’s key people, fees, redemption periods, restrictions on leverage, disclosure
policies), annual reports, marketing material, hedge fund databases and simple ques-
tionnaires that are sent to the fund management.420 The focus of qualitative analysis
at this stage is to obtain a deeper understanding of a hedge fund’s returns and the
fund’s investment approach in general. In order to determine the comparative ad-
vantage of a hedge fund manager, potential investors usually search for one of the
following reasons: better information, better analysis of information, better portfo-
lio implementation, and/or (particularly in private markets) access to proprietary
investments.421 Furthermore the quantitative side is supplemented by analyzing the
implementation of trades and by ensuring the validity of the track record.422
The result of the combination of investor’s views and the qualitative and quantitative
analysis is the so-called short list. The remaining managers on this short list are
contacted and monitored on a regular basis but are at first subject to the heart of
the hedge fund selection process: the due diligence. The due diligence process
can be described as a more comprehensive and more thorough form of a qualitative
analysis. The sources of information in such a detailed analysis are due diligence
questionnaires, visits to the hedge fund’s offices and interviews with the management
team.423 At this point the fund of funds manager tries to rule out most of the
industry-specific risk factors outlined in Section 3.1.1.
The qualitative due diligence process can be structured in many different ways.
Figure 3.5 distinguishes between four main fields for the due diligence.
As a starting point the manager due diligence takes a look at the investment process
and the strategy of the fund. The fund itself has to be considered as well and the
420 See Jaeger (2002), p. 239, and Lhabitant (2002), p. 189.421 See Spitz (2001), p. 61.422 See Lhabitant (2002), p. 188.423 See Lhabitant (2002), p. 189.424 See Jaeger (2002), pp. 237-246.
3 Hedge Fund Strategies 102
��������������� ������������������
����������������
� �������������
���������������
�� ����������������
���
���������
��������
�������������
������������������������
����������
���
� ��������� ���
��������������������
�� ����������� �������
��������������� ��������
!������
���
����� ��������
�������������
�� �"���!��"��
���������������
���
Figure 3.5: Important topics in a hedge fund due diligence424
analysis of the fund’s legal structure is another important point. Here the terms
of the hedge fund are in the focus of the due diligence. A sound risk management
system might play a vital role for the hedge fund investment, therefore the portfolio
constraints and the risk monitoring are also subject to thorough investigation in
the process of a due diligence. Finally the infrastructure of the fund, a more tech-
nical matter, is also examined. The investor should assure that the implemented
infrastructure does not impose any constraints on the fund’s development.425
The results of the due diligence are usually aggregated in a scoring system which
allows the ranking of funds on the short list. A formal report should document
the resulting decision to invest or not to invest or certain recommendations.426 The
due diligence information is kept up to date with monthly performance reviews and
425 See for example Jaeger (2002), pp. 237-246, for a very detailed description of the due diligence
process.426 See Lhabitant (2002), p. 192.
3 Hedge Fund Strategies 103
management contacts. Usually the broad-based due diligence process is updated
every year.427 If a fund has been selected and added to the portfolio, an ongoing
monitoring process should be started. This permanent due diligence should support
the decision making on the question wether the investment should be left unchanged,
more capital should be committed or assets in whole or part should be redeemed.428
3.4 Summary
In this chapter a broad range of different hedge fund strategies was introduced. For
a useful analysis of hedge fund strategies an investor has to distinguish industry-
inherent and strategy-specific risk factors. While almost every hedge fund bears
at least some of the industry-inherent risk factors, the risk profile of hedge funds
considering strategy-specific risks is very different. Therefore, we outlined the most
important trading strategies and the corresponding range of risk factors for a broad
variety of strategies.
Furthermore, funds of hedge funds were introduced. These funds allow investors
to profit from an already diversified portfolio of several single hedge funds. We
described the most important advantages and disadvantages of these investment
vehicles and illustrated a typical fund selection process.
Altogether, this chapter was focused on qualitative aspects of hedge fund invest-
ments. In order to make some statements on hedge funds that are based on actual
data the next chapter will take a look at the world of hedge fund indices.
427 See Perisse (2004), p. 15.428 See Lhabitant (2002), p. 193.
4 Description and Analysis of Hedge
Fund Data
This chapter gives a general review on hedge fund data. At first most of the different
database providers and the corresponding hedge fund indices are introduced. Most
of the hedge fund index universe is described in detail in this section. Biases in
the resulting time series are the object of Section 4.2. As hedge fund databases
will never be able to mirror the whole hedge fund population, we have to consider
systematic biases in hedge fund data. As these biases will affect the indices that are
calculated on database samples.
After the description of data bases and the corresponding problems with biases
we will focus on the analysis of some of the resulting index series. Because of
different problems and biases fund of hedge fund data will be used to proxy the
performance of hedge fund investments. In Section 4.4 the statistical properties of
fund of hedge fund index data including unconditional return distributions and time
series behaviour will be examined in greater detail.
4.1 Hedge Fund Databases and Index Providers
In the following the main providers of hedge fund databases and indices are described
in more detail. The next section gives a brief overview on desirable properties of
104
4 Description and Analysis of Hedge Fund Data 105
hedge fund indices while Sections 4.1.2 and 4.1.3 contribute information on the index
providers and the index methodology, respectively.1
4.1.1 Desirable Properties of Hedge Fund Indices
In this section we will briefly describe the usage of hedge fund indices and desirable
properties for these indices. Hedge fund indices serve as benchmarks for hedge fund
investments.2 Only if such a benchmark for relative comparisons is available, the
risk adjusted alpha of a hedge fund investment can be determined.3 Therefore,
hedge fund indices provide a service in measuring performance. In order to allow
“passive” investments in this part of the alternative investment universe, indexing
is also necessary. With diversified index products investors could achieve a broad-
based, efficient and (relatively) low cost hedge fund exposure.4 Furthermore indices
clarify hedge fund strategies and the corresponding risk and return characteristics.
In general indices are able to enhance the hedge fund industry’s transparency.5
There is a variety of desirable properties for hedge fund indices.6 The construction
of hedge fund indices should consider most of these competing objectives in order
to obtain a useful benchmark. A summary over these benchmark objectives and the
corresponding problems within the hedge fund industry is found in Figure 4.1 on
the following page.
A first desirable attribute for a hedge fund benchmark is the completeness of such
1 The information on hedge fund database providers and the calculated indices used in the follow-
ing sections was collected from the internet, official documents and company representatives.2 Other market indices (like for example stock market indices) are not able to capture the nature
of hedge fund investing.3 See for example Zask et al. (2004).4 See Kohler (2003), p. 5.5 See Zask et al. (2004).6 See for example Ruckstuhl et al. (2004) for an overview on construction requirements.7 See Kohler (2003), p. 8.
4 Description and Analysis of Hedge Fund Data 106
���������� ������� ����������������������������
������������������
��������
�������������
�����������
��������������� �� �
���� ���������� ������������
�������������
��� ��� ����������� ������������
����������� ����������������
���������� ������ ��� ����������
����� ���������� �������� �
��� ����� �� ��!��� ������ �
"�������������!� ���� �
#���������� ������������
��$�� �����!� ���� ������ ���
Figure 4.1: Objectives and problems of hedge fund benchmark construction7
an index. The index should represent the hedge fund universe accurately. In the
index all of the available instruments within the hedge fund industry should be rep-
resented directly or the index should at least approximate them. This full coverage
is essential for hedge fund indices. There are various problems with this demand
for representativeness. At first there is a self-reporting bias in hedge fund data, as
some managers will not submit data to index providers.8 Multi-strategy funds are
also a problem, as they are difficult to classify in the set of available hedge fund
strategy groups. So they are often not included in the index-calculations. Finally
a representative index should comprise hedge funds that are open to new investors
and closed hedge funds that do not accept further capital. These closed funds are
still active hedge funds but they are not investable and therefore they violate the
investability objective of hedge fund benchmark construction.9
8 In Amenc/Martellini (2002) the authors estimate that only half of the existing hedge funds
report their performance data to at least one major index provider, see Amenc/Martellini
(2002), p. 5.
4 Description and Analysis of Hedge Fund Data 107
As a second attribute of a good index, the accuracy of the calculated index is very
important. The price discovery process of the index should be reliable, consistent
and verifiable by an independent third party. For the underlying hedge fund data
the same holds true. This performance information should be readily available and
reliable. This is a problematic point in the hedge fund industry where most managers
report unaudited data to hedge fund database providers.10
The calculated indices should be transparent in order to be easily replicable. There-
fore, the index should be well and unambiguously defined, and the index construc-
tion should follow clear and objective guidelines that are known in advance. When
hedge fund indices are calculated, the index construction methodology is not always
well-defined and articulated. Often index committees are involved in the selection of
index funds and the underlying strategies are not always unambiguously classified.11
A useful benchmark should be investable. Investors should be able to achieve a
benchmark exposure easily. Therefore, the indices should be designed to facilitate
product development.12 The indices should represent a well-defined (systematic) risk
premium that is available through passive investing. As mentioned above a main
problem with this investability objective is the resulting exclusion of closed funds
from the index calculation.13 A problem with open funds might be their limited
capacity to accept new investors. So the benchmark (or some constituents) might
become un-investable when funds reach their capacity limits. Besides these capacity
constraints the limited liquidity of hedge fund investments in general (concerning
minimum investment limits, lock-up and redemption periods) is problematic.14
9 See Kohler (2003), p. 8, and Jaeger (2004).10 See Kohler (2003), p. 8, and Jaeger (2004).11 See Kohler (2003), p. 8, and Jaeger (2004).12 See Patel (2003).13 For the trade-off between representation and investability see for example Patel et al. (2003),
p. 62.14 See Kohler (2003), p. 8, Jaeger (2004), and Section 2.3.6 on page 20.
4 Description and Analysis of Hedge Fund Data 108
As we see, these objectives are partially competing and in the hedge fund industry
not all of them can be met at the same time. Therefore, the calculated indices
always represent a trade-off between the various objectives.15,16
4.1.2 Hedge Fund Index Providers
In this section 21 hedge fund index providers are described in detail. We will take
a closer look at the corresponding hedge fund databases and the particular index
methodologies.
Alternative Asset Center
The Alternative Asset Center (AAC) provides fund of hedge fund data.17 The index
offered by AAC is calculated since 2001 and offers a history beginning in 1996. The
equal weighted index is published monthly with a time lag of 5 to 8 weeks. The
indices are not investable that means the funds in the database might be closed
to new investors. The company claims to operate the largest fund of hedge funds
database in the world. This database currently features information on 1700 on-
and offshore funds of hedge funds. All of these funds are included in the published
index.
ABN AMRO Eurekahedge Index
ABN AMRO and Eurekahedge (ABN/EH) calculate an index series since 2002 with
a history that dates back to 1999.18 The index series is focused on Asian hedge funds
and offers a composite and two additional regional indices. The Asian hedge fund
15 See for example Patel (2003).16 For a list of alternative strategy specific hedge fund benchmarks, see for example Jaeger (2004).17 See www.aa-center.net.18 See www.eurekahedge.com.
4 Description and Analysis of Hedge Fund Data 109
database from Eurekahedge currently comprises 468 funds and the non-investable
indices represent 147 hedge funds. The final equal weighted index values are calcu-
lated with a time lag of two months.
Altvest/InvestorForce
The database provider Altvest started in 1996 to publish its hedge fund indices.19
In 1999 it was acquired by InvestorForce. The original Altvest indices offered data
going back to 1993 but the Altvest sub-indices prior to the year 2000 had to be
recalculated due to biases. The monthly values are published from the beginning of
the following month on and are permanently updated for the data of funds reporting
their recent figures. On the last day of the following month the final index values
are calculated. The indices are not investable. The database comprises about 2600
hedge funds and the composite index combines 2245 of these funds. Besides this
composite index, 13 strategy indices are calculated. In all of the indices the hedge
funds are equally weighted.
Barclay Trading Group/Global HedgeSource
Global HedgeSource was founded in 2002 by Barclay Trading Group.20 The first
indices were calculated in 2003 and the history was filled until 1997. A total of 18
hedge fund indices and additionally a top ten fund ranking for all the strategies is
published every month. Index estimates are released the next or the next but one
month and the final index values are released three months after. The database
consists of about 3000 funds and all of these funds are included in the index calcu-
lations for one composite, one fund of funds and 16 strategy indices. All of these
indices are equally weighted.
19 See www.investorforce.com.20 See www.barclaygrp.com.
4 Description and Analysis of Hedge Fund Data 110
Bernheim Index
Since 1995 Dome Capital Management publishes the so-called Bernheim Index.21
From 1995 to 1999 the index was calculated on a quarterly basis and since 1999
it is updated monthly. The index values are available with a time lag of 2 to 5
weeks. The published index is based on the well-known US Offshore-Funds Directory
(OFD) which includes about 1000 offshore hedge funds but only 18 representative
(“leading”) fund managers are included in the equally weighted Bernheim Index.
Besides this composite index there are no strategy indices available.
BlueX
The investable BlueX (Blue Chip Hedge Fund Index) is the result of a cooperation
between the Austrian Stock Exchange and Benchmark Capital Management.22 The
index started in July 2002 and is calculated back to December 2001. The Benchmark
database comprises 2500 hedge funds but only 30 to 40 (blue chip) hedge funds
that are managed and controlled by large investment banks or leading financial
organizations are included in the BlueX index. The estimated index is calculated on
a weekly basis and the final index is determined on a monthly basis with a maximum
delay of 25 days.23 The calculation scheme of the index is a mix of value and equal
weighting of the individual funds.
CISDM/MAR
The Managed Accounts Reports (MAR) index series was established in 1994 and the
calculated index values started in 1990.24 The hedge fund index series was then sold
21 See www.hedgefundnews.com.22 See www.bluex.org and Benchmark Advisory (2004).23 When the final monthly index value is calculated a maximum of 15% estimated fund values is
used to determine the BlueX.
4 Description and Analysis of Hedge Fund Data 111
to Zurich Capital Markets and since August 2002 the hedge fund indices are calcu-
lated by the Center for International Securities and Derivatives Markets (CISDM)
at the University of Massachusetts.25 The database which is still maintained by
MAR contains about 1200 single hedge funds and about 500 fund of hedge funds.
Additionally the database includes 1500 inactive funds. A total of 20 indices is
calculated from all active hedge funds. There are 9 strategy indices and 11 corre-
sponding sub-indices that represent the median of the funds in the strategy groups.
The monthly indices are calculated with a time lag of two months.
CSFB/Tremont
The joint venture CSFB/Tremont was initiated by Credit Suisse First Boston and
Tremont Capital Management.26 CSFB/Tremont was established in 1999 and the
index history dates back to 1994. The monthly indices are value weighted and
published on the 15th of the following month. The funds are taken from the Tremont
TASS database with data on over 3000 hedge funds managed by more than 1300
managers.27 The number of hedge funds from the TASS database that is used for
the calculation of the composite index and the 10 sub-indices is about 430.
Since August 2003 CSFB/Tremont also offers a series of investable indices. The
CSFB/Tremont Investable Hedge Fund Index (CSFB inv) is calculated back to the
year 1999 and it is investable through an index-tracker fund.28 The index is based
on the funds for the common CSFB/Tremont index series. From these funds (a total
of about 430) 60 funds are selected for the investable index. The investable index
series comprises 10 sector indices. From every sub-index category the 6 largest funds
24 See www.marhegde.com.25 See http://cisdm.som.umass.edu.26 See www.hedgeindex.com, CSFB/Tremont (2002), CSFB/Tremont (2004a) and
CSFB/Tremont (2004b).27 See www.tremont.com.28 See for example Goricki/Sohnholz (2004), p. 34, for a brief discussion of index-tracker funds.
4 Description and Analysis of Hedge Fund Data 112
that meet certain conditions are included in the Investable Hedge Fund Index.29 US
domiciled hedge funds are not part of the investable funds universe.
Dow Jones Hedge Fund Indices
Dow Jones Hedge Fund Indices is a relatively new name in the hedge fund industry.30
Since 2004 an investable hedge fund index series is calculated that is based on a
managed accounts platform. This series comprises five strategy indices, the so-
called Dow Jones Hedge Fund Strategy Benchmarks, and a portfolio index that is
allocated in these strategies, the Dow Jones Hedge Fund Balanced Portfolio Index31.
The monthly values of the equal weighted strategy indices start at the end of 200132
and daily index values are available since the beginning of 2004. For the Balanced
Portfolio Index daily and weekly data is provided. The final values of the indices are
published with a time lag of two months. In order to select the currently 35 funds
for the index series Dow Jones obtains fund information from different information
providers.
Evaluation Associates Capital Markets
Evaluation Associates Capital Markets (EACM) calculates hedge fund indices since
1996.33 The EACM100 Index which only comprises US hedge funds was launched
in 1996 and was originally calculated back to the year 1990. But due to biases the
29 See CSFB/Tremont (2004a) for more details on eligible funds for the CSFB/Tremont Investable
Hedge Fund Index.30 See www.djindexes.com, Dow Jones Indexes (2004a), Dow Jones Indexes (2004c) and Dow
Jones Indexes (2004b).31 In fact there are two indices, the Dow Jones Hedge Fund Balanced Portfolio Index AX and BX,
that reflect the development of a hedge fund portfolio. They differ as the Dow Jones Hedge
Fund Balanced Portfolio Index BX documents the performance of an investment that is hedged
in Euro.32 Data on the strategy index “Equity Market Neutral” is only available since 2003.33 See www.eacm.com.
4 Description and Analysis of Hedge Fund Data 113
index data before 1996 is no longer published. Since 2004 Evaluation Associates
Capital Markets also offers the “EACM100 Index - Offshore Funds” which focuses
only on offshore hedge funds. For this index no historical performance information
is available before the year 2004. The values of both equal weighted indices are pub-
lished with one month delay and they are based on 100 hedge funds that have been
directly selected. Fund managers are grouped into 5 broad investment strategies,
which are further cut into 12 sub-strategies.
EDHEC
Since march 2003 EDHEC, a French business school, calculates a series of hedge
fund indices that is different from other index concepts in the industry.34 As sug-
gested by Amenc/Martellini (2002), EDHEC publishes a series of “indices of in-
dices” because the resulting index series is less biased and more representative for
the hedge fund universe. Therefore, the virtual EDHEC database is the sum of
the databases of other index providers. Namely the indices of Altvest, Barclay,
CISDM, CSFB/Tremont, EACM, Hennessee Group, HEDGEFUND.NET, HFR,
Standard&Poors and VAN are currently used in the EDHEC calculations. The
weighting of these different indices is determined by principal components analysis
(PCA). The published index series with 13 strategy indices starts in 1997 and the
final monthly values are available with a delay of one month.
Feri Trust
The Absolute Return Investable Index (ARIX) series that is published by Feri Trust
is calculated since 2003 and the index history dates back to 2002.35 The monthly
equal weighted36 indices are published with a time lag of up to 6 weeks. The Feri
34 See www.edhec-risk.org, Amenc/Martellini (2002) and EDHEC (2004).35 See www.feritrust.de.
4 Description and Analysis of Hedge Fund Data 114
database is a collection of TASS, VanHedge and HF Net data and additional funds.
According to Feri there are currently more than 6000 funds and managers included.
41 hedge funds form the ARIX composite index. The ARIX series offers 2 master
and 4 strategy indices.
FTSE Hedge Index
FTSE publishes a broad range of financial indices and since 2004 the FTSE Hedge
Index is calculated with a history beginning in 1997.37 The FTSE composite index
is made up of 3 style indices which can be further divided into 8 strategy indices.
The composite index includes a total of 40 hedge funds. The monthly final values
of FTSE indices are published with a delay of about six weeks. Additionally FTSE
offers the daily indices with a time lag of 3 days. The weighting of the funds in
these indices is done according to the fund’s investablility measured as remaining
fund capacity. The FTSE database is collected from different hedge fund database
providers and contains additional FTSE data.
Hennessee Group
The Hennessee Group has a long tradition calculates hedge fund indices.38 Since
1987 Hennessee Group computes hedge fund indices and currently publishes an
index history beginning in 1993. A total of 24 indices is calculated, a composite
index (Hennessee Hedge Fund Index) and 23 strategy indices. Index values are
published with a two week delay but final index values are made public not until
the beginning of the following year. The Hennessee database contains 3000 on- and
offshore funds and 750 of these funds are included in the index calculations.
36 The ARIX composite index is also calculated as value weighted index but this index is not
published.37 See www.ftse.com, FTSE (2004a) and FTSE (2004b).38 See www.hennesseegroup.com.
4 Description and Analysis of Hedge Fund Data 115
Hedge Fund Intelligence
London-based Hedge Fund Intelligence (HFI) currently determines 3 index series,
the EuroHedge-, the AsiaHedge- and the InvestHedge-Index with different numbers
of sub-strategies.39 These indices were first published in 2001, 2002, and 2003 and
are all calculated back to 1998. The new index values are distributed the next month.
The database for the EuroHedge (EH) series comprises 1000 hedge funds, for the
AsiaHedge (AH) series about 400 funds and the InvestHedge (IH) series database
includes 1300 (fund of) hedge funds. In the course of the actual index calculation
850, 290 and 900 funds are used when the median values are determined.
HEDGEFUND.NET
Another source for hedge fund indices is the online provider HEDGEFUND.NET
(HF-NET) which offers the so-called Tuna index series.40 Founded in 1997 HEDGE-
FUND.NET started calculating indices in 1998 and offers a history for some series
that dates back to 1976. HEDGEFUND.NET monthly publishes 5 composite in-
dices and a broad range of 33 strategy indices that are released at the beginning of
the next month. As the index values are permanently updated, there are no final
index values available. The equal weighted Tuna index calculations are based on
data from more than 3900 hedge funds that report to HEDGEFUND.NET. All of
these funds are included in the published indices.
Hedge Fund Research
Hedge Fund Research (HFR) offers a well-established index series.41 The monthly
HFR indices (HFRI) are calculated since 1994 with an index history beginning in
39 See www.hedgefundintelligence.com.40 See www.hedgefund.net.41 See www.hfr.com.
4 Description and Analysis of Hedge Fund Data 116
1990. The individual hedge funds in these indices are equally weighted. In 2003
HFR launched a series of daily indices that are investable. These HFRX indices
offer a history beginning in the year 2000 and the index values are published the
next day. For the HFRI series the final index values are made public with a time
lag of one month. Hedge Fund Research works with a database of about 4000 hedge
funds and 1650 of these funds are included in the calculation of the HFRI series.
HFRX indices are calculated from the same basic database but no information on
the number of funds used in the averaging is given. Hedge Fund Research publishes
36 strategy indices for the HFRI and 8 strategy indices for the HFRX.
Mondo Hedge Index
The Italian Mondo Hedge Index was created in 2003 and offers an index history
beginning in 2001.42 The focus of Mondo Hedge is to provide a benchmark for
Italian fund of hedge fund managers. Therefore, data from 107 Italian funds of
hedge funds is currently included in the database for the Mondo Hedge Index series
and the results of 102 funds are used in the process of index calculation. The
index series that is published with a time lag of about five weeks offers a composite
fund of hedge funds index and two strategy indices with a total of four sub-indices.
The whole index series is calculated on the basis of a value-weighting and an equal
weighting scheme.
MSCI Hedge Fund Indices
In 2002 Morgan Stanley Capital International (MSCI) started to calculate a hedge
fund index series that covers many hedge fund strategies with individual sub-indices.43
A total of 190 hedge fund indices are calculated and published in different strategy
42 See www.mondohedge.com.43 See www.msci.com, MSCI (2002), MSCI (2003b) and MSCI (2003a).
4 Description and Analysis of Hedge Fund Data 117
groups. The composite index is determined on an equal and a value weighted basis
and has an index history until 1994. The indices are based on data from about 1800
hedge funds. About 98% of these funds are included in the index calculations. The
final index values are made public with a delay of one month.
In 2003 MSCI in cooperation with Lyxor Asset Management released a series of
investable hedge fund indices (MSCI inv). A history of index values has been estab-
lished that dates back to the year 2000. Based on data of currently 97 hedge funds
that is supplied by Lyxor Asset Management weekly index values are published.
Standard & Poors Hedge Fund Indices
The well-known index provider Standard & Poors (S&P) started to publish hedge
fund indices in 2002.44 Index values are available from 1998 on. The Standard &
Poors index series is calculated on a daily basis and published with a time lag of two
days. The official monthly indices are released after 3-4 weeks. The database for
the investable Standard & Poors index series comprises 3500 funds from proprietary
Standard & Poors data and other data providers. For the index calculation 40 hedge
funds are included.
VAN Hedge Fund Indices
VAN Hedge Fund Advisor was founded in 1988 and started in 1994 to calculate the
VAN hedge fund index series that was published for the first time in 1995.45 The
three monthly index series Global, US-Onshore and Offshore date back to 1988 and
the corresponding indices are calculated with an equal weighting scheme. The final
index values are published with a delay of up to one month. The VAN database
44 See www.standardandpoors.com, Standard & Poor’s (2003b), Standard & Poor’s (2003a), and
Standard & Poor’s (2004).45 See www.vanhedge.com.
4 Description and Analysis of Hedge Fund Data 118
comprises more than 6000 hedge funds whereof about 1000 funds are no longer
active. The final index values consider more than 1000 funds. The index-tracker
fund that makes the Offshore-Index investable comprises about 45 hedge funds.
4.1.3 Hedge Fund Databases and Index Methodologies
For the previously mentioned index and database providers some basic information is
summarized in Table 4.1 on the following page. The summary includes information
on the launch of the index series and the historical index data. Furthermore the
weighting scheme of the indices is considered. The different calculation methods
employed are equal weighting (ew), value weighting (vw), an implicit weighting
scheme based on principal components analysis (PCA),46 capacity weighting (cw),47
and the usage of the statistical median. In an equal weighted index the index
values are determined by averaging over a number of funds, that means every hedge
fund gets the same weight, while a value weighted index reflects the assets under
management (AUM) of the funds in the corresponding fund weights.48
Whether an index series is investable or not depends on the hedge funds included in
the corresponding database. The column “Investable” contains information wether
investors are able to place money in the underlying portfolio or index derivatives.
according to the index providers. Finally the last column of Table 4.1 on the follow-
ing page gives a short summary on the time lag when performance data of hedge
fund indices is published.
The main strategy indices that are available from the different index providers are
46 See Amenc/Martellini (2002), pp. 13-24, for a description of the PCA-weighting scheme for
EDHEC indices.47 See FTSE (2004b), pp. 8 and 10, for a description of the capacity weighting scheme for FTSE
indices.48 Investment strategies that are implied by different weighting schemes are outlined in Sec-
tion 4.2.2 on page 139.
4 Description and Analysis of Hedge Fund Data 119
Start History Weight- Invest- Published
ing able finally after
AAC 2001 1996 ew no 5-8 weeks
ABN/EH 2002 1999 ew no two months
Altvest/IF 1996 1993 ew no one month
Barclay 2003 1997 ew no three months
Bernheim 1995 1995 ew no 2-5 weeks
BlueX 2002 2001 ew/vw yes two months
CISDM 1994 1990 median no two months
CSFB 1999 1994 vw no two weeks
CSFB inv 2003 1999 vw yes six weeks
Dow Jones 2004 2001/2003 ew yes two months
EACM 1996/2004 1996/2004 ew yes 2-3 weeks
EDHEC 2003 1997 PCA no one month
Feri 2003 2002 ew yes six weeks
FTSE 2004 1997 cw yes six weeks
Hennessee 1987 1993 ew no year-end
HFI 2001-2003 1998 median no one month
HF-NET 1998 1976 ew no —
HFRI 1994 1990 ew no one month
HFRX 2003 2000 ew/vw yes one week
Mondo 2003 2001 ew&vw no five weeks
MSCI 2002 1994 ew/vw no one month
MSCI inv 2003 2000 ew yes one week
S&P 2002 1998 ew yes 3-4 weeks
VAN 1994 1988 ew no/yes one month
Table 4.1: Summary on hedge fund indices and index providers
4 Description and Analysis of Hedge Fund Data 120
presented in Table 4.2 on the next page. Only the most common hedge fund strate-
gies with a minimum of four representatives from different index providers have
been listed. The listing starts with the composite indices and the fund of funds
indices. The 12 individual hedge fund strategies considered in Table 4.2 are: eq-
uity market neutral, long/short equity, convertible arbitrage, fixed income arbitrage,
relative value, merger arbitrage, distressed securities, event driven, global macro ,
short selling, emerging markets, and managed futures.49
After examining the databases and the calculated strategy indices we will take a
closer look at the objectives for benchmark construction that have been outlined in
Section 4.1.1.
At first the representativeness of the different hedge fund databases and indices is
examined. Table 4.4 on page 123 gives a summary on the representativeness of the
hedge fund databases used in the course of index calculation. Here the number of
funds in the databases, the number of funds in the resulting index series and the
restrictions for the acceptance of funds in the database are outlined.50 Table 4.5
on page 124 offers information on the rules for index construction. The selection
of funds from the database by the corresponding index providers is examined ac-
cording to whether a minimum track record or a minimum amount of assets under
management is required and if closed funds are included in the index calculation.
Furthermore, Table 4.5 outlines if the strategy categorization is done by the hedge
fund manager or the index provider.
The accuracy of the calculated indices with the problem of unreliable and/or unau-
dited fund performance is addressed in the column “Verification of data” in Table 4.6
on page 125. Furthermore the transparency of the index calculation is analyzed in
Table 4.6. Here we examine wether the index selection criteria, the index calculation
49 See Chapter 3 for more information on these hedge fund strategies.50 The number of funds in the different databases that is reported in Table 4.4 on page 123 is
from September and October 2004.
4 Description and Analysis of Hedge Fund Data 121
Com- Fund of Equity Long Conv. Fixed Rel.
posite Hedge Market Short Arb. Inco. Value
Funds Neutral Equ. Arb.
AAC X
ABN/EH X
Altvest/IF X X X X
Barclay X X X X X X
Bernheim X
BlueX X
CISDM X X
CSFB X X X X X
CSFB inv X X X X X
Dow Jones X X X
EACM X X X X
EDHEC X X X X X X
Feri X X
FTSE X X X
Hennessee X X X X
HFI EH X X X
HFI AH X X
HFI IH X X X X
HF-NET X X X X X X X
HFRI X X X X X X X
HFRX X X X X
Mondo X X
MSCI X X X X
MSCI inv X X X
S&P X X X X X X
VAN X X X X X X
Table 4.2: Summary on hedge fund index strategies I
4 Description and Analysis of Hedge Fund Data 122
Merger Distr. Event Global Short Emerg. MF
Arb. Sec. Driven Macro Selling Markets
AAC
ABN/EH
Altvest/IF X X X X X X
Barclay X X X X X X
Bernheim
BlueX
CISDM X X X X X X
CSFB X X X X X
CSFB inv X X X X X
Dow Jones X X X
EACM X X X
EDHEC X X X X X X X
Feri X
FTSE X X X X
Hennessee X X X X X
HFI EH X X X
HFI AH
HFI IH
HF-NET X X X X X X X
HFRI X X X X X X
HFRX X X X X
Mondo
MSCI X X X X
MSCI inv X X
S&P X X X X X
VAN X X X X X
Table 4.3: Summary on hedge fund index strategies II
4 Description and Analysis of Hedge Fund Data 123
Database Funds in Funds in Restrictions
database Indices
AAC Own data 1700 all —
ABN/EH Own data 462 147 Asian funds only
Altvest/IF Own data 2600 2200-2300 —
Barclay Own data >3000 all —
Bernheim Offshore
Funds Direc-
tory (OFD)
1047 18 Offshore funds only
BlueX Own data 2500 33 —
CISDM MAR ≈ 1800 all —
CSFB TASS 3000 433 —
CSFB inv TASS 3000 60 Offshore funds only
Dow Jones MAR, HFR,
OFD
sum(overlap) 35 —
EACM Own data 100 100 Offshore funds only,
onshore funds only
EDHEC Sum of different databases and indices —
Feri TASS, VAN,
HF-NET, and
other sources
6000 41 No US-Onshore
funds
FTSE Own data and
other sources
6000 40 —
Hennessee Own data 3000 750 —
HFI Own data
and Bank of
Bermuda
1000/400/1300 850/290/900 European, Asian,
Global funds
HF-NET Own data 3900 all —
HFRI Own data 4000 1650 —
HFRX Own data 4000 — —
Mondo Own data 107 102 Italian funds only
MSCI Own data 1800 ≈ 98% —
MSCI inv Lyxor Asset
Management
97 —
S&P Own data and
other sources
3500 40 —
VAN Own data 6000 >1000 —
Table 4.4: Summary on hedge fund database representativeness
4 Description and Analysis of Hedge Fund Data 124
Minimum Minimum Closed Classification
History AUM in USD Funds done by
AAC — — yes —
ABN/EH — 40 million no index provider
Altvest/IF — — yes fund manager
Barclay — — yes index provider
Bernheim — — yes —
BlueX — 20 million no —
CISDM — — yes fund manager
CSFB 1 year 10 million yes index provider &
fund manager
CSFB inv 1 year 10 million no index provider &
fund manager
Dow Jones 2 years 50 million no index provider
EACM 2 years 20 million no index provider
Feri 1-3 years 50 million no index provider
FTSE 2 years 50 million no index provider
Hennessee 1 year 10 million yes —
HFI — — yes index provider
HF-NET — — yes fund manager
HFRI — — yes fund manager
HFRX — 5 million no fund manager
Mondo — — yes index provider
MSCI — — yes index provider &
fund manager
MSCI inv — yes no index provider &
fund manager
S&P 3 years 75 million no index provider
VAN 3 months — yes index provider
Table 4.5: Summary on hedge fund index representativeness
4 Description and Analysis of Hedge Fund Data 125
Selection Calculation Index Veri- Index
criteria method constit- fication commit-
details uents of data tee
AAC — unknown unknown — yes
ABN/EH known unknown unknown no yes
Altvest/IF known known unknown yes no
Barclay — known known yes yes
Bernheim unknown unknown known yes no
BlueX known known known — yes
CISDM — unknown unknown yes no
CSFB known known known yes yes
CSFB inv known known known no yes
Dow Jones known known known yes no
EACM known unknown unknown no no
EDHEC known known — —
Feri unknown unknown unknown yes yes
FTSE known known known yes yes
Hennessee known unknown unknown yes(finals) no
HFI known unknown unknown yes yes
HF-NET — unknown known no no
HFRI known unknown unknown no no
HFRX unknown unknown unknown yes —
Mondo known known known — yes
MSCI known known unknown yes yes
MSCI inv known known unknown yes yes
S&P known known known yes yes
VAN known unknown unknown no no
Table 4.6: Summary on hedge fund index transparency
4 Description and Analysis of Hedge Fund Data 126
methods and/or the index constituents are published. Another problem regarding
the transparency of hedge fund indices is the use of an index committee to revise the
index composition. For the introduced index providers the summary on these index
committees can be found in the last column of Table 4.6 on the preceding page.
4.2 Biases in Hedge Fund Index Data
After addressing the problems with hedge fund index methodologies in Section 4.1
we will now focus on different biases in hedge fund index data. The data included
in different databases represents only subsets of the hedge fund universe and the
measurement problems that result from this incompleteness of hedge fund data
have to be considered when fund returns are analyzed and interpreted.
In the next sections systematic biases in hedge fund data are explained and the
consequences of these findings for hedge fund analysis are documented. In Section
4.2.1 we describe the well known biases in hedge fund databases, and subsequently
the problems with different weighting schemes for hedge fund index calculation are
outlined in Section 4.2.2. As a strategy index group that is able to cope with at
least some of these problems we will finally introduce fund of hedge funds in Section
4.2.3.
4.2.1 Biases in Hedge Fund Databases
Unfortunately hedge fund databases are exposed to a number of biases. As already
outlined, there is no central data pool for hedge funds where fund managers have
to report their figures to.51 Therefore, the whole population of hedge funds and
51 As hedge funds are frequently organized as offshore or private investment funds, they are usually
not even enforced to disclose their results to the public, see for example Fung/Hsieh (2002a),
p. 23.
4 Description and Analysis of Hedge Fund Data 127
corresponding hedge fund return series is unknown. As a consequence the databases
are not fully representative for the hedge fund population as a whole and samples
from the database will give approximations and biased measures for the performance
of a hedge fund investor. We will discuss the most important biases that are located
on the fund level and on the level of the database provider, respectively.
At first we will take a look at the reasons and explanations for the different biases.
Then we will examine the nature of four biases in greater detail. We will describe
survivorship bias, selection bias, instant-history bias and the stale-price bias. Finally
the consequences for the estimation of return characteristics like expected return,
standard deviation and higher moments from hedge fund databases are discussed in
Section 4.2.1.6.
4.2.1.1 Problems with Hedge Fund Data
There are different characteristics of the hedge fund industry that entail at least
some of the biases outlined in Sections 4.2.1.2 to 4.2.1.5. These biases originate
from problems on different levels of the information flow. We will focus on problems
and the resulting data biases that are a direct result of the data collection process.
The considered biases are therefore usually caused by problems on the level of hedge
funds and on the database level. Figure 4.2 on the next page gives an overview on
some problems and the resulting biases. The researcher which might be another
source of biases is excluded in the further analysis.
To start with the origin of hedge fund data we will take a closer look at hedge
funds themselves. Here the reporting of the data is in the focus of our analysis. As
the reporting to database providers is voluntary and the reported net asset values
or return figures are usually subject to at least slight adjustments by hedge fund
managers, the performance of funds in a database might differ from the actual
52 See Signer (2002), p. 29, and Signer (2003), p. 69, for related figures.
4 Description and Analysis of Hedge Fund Data 128
����������
����������
��� ����� ���� ���
������ ������
������������������
��� ���
�����
������������������
����������������
������������������
������������������
�����������������
������������
��������� �����
��� ���������
�����������
��������!���������
�������������
Figure 4.2: Problems with hedge fund data and resulting biases52
performance of the hedge fund population.53
On the level of data providers further problems that result in biases are the be-
ginning of data collection in the mid 1990s, the backfilling of historical data, the
inclusion of dead funds in the historical database and the different criteria or mini-
mum requirements for including hedge funds into the database.54 These reasons for
data problems are outlined in the following sections when the corresponding biases
are discussed.
4.2.1.2 Survivorship Bias
Survivorship bias55 does arise when a sample of hedge funds includes only funds
53 See for example Signer (2002), pp. 29-30.54 See for example Signer (2002), p. 29.
4 Description and Analysis of Hedge Fund Data 129
that are alive and operating at the end of the sample period.56 This bias does occur
because of problems with hedge fund databases or because the researcher starts with
a list of funds that exist at the end of the analyzed sample period.57 The part of the
survivorship bias that is influenced by the researcher58 is excluded in the following
discussion as we focus on database problems and biases.
A first problem that leads to survivorship bias is located in the history of hedge fund
databases. For funds that ceased to exist before the start of systematic collection of
hedge fund data in the 1990s no extensive dataset is available because these funds
predated the existence of most hedge fund databases. Therefore, historical hedge
fund data over this time frame is usually based on funds that survived until this
early period which leads to survivorship bias. The hedge fund data recorded in
these years will usually overestimate the returns for the hedge fund population as a
whole.59
Another source of survivorship bias is found in the exclusion of funds that have
stopped reporting to database providers. These funds from the hedge fund popula-
tion are no longer part of the observable database subset. Here we can differentiate
funds that ceased operations, that voluntarily stopped reporting and funds that were
delisted by the database provider. The funds that stopped reporting and left the
database are usually called “defunct” funds as opposed to surviving funds that are
still alive and reporting to the database providers.60 These different ways to exit
55 The “survivorship bias” is also called “survivor bias”, see for example Ackermann et al. (1999),
p. 864, Schneeweis et al. (2001), p. 20, and Howell (2001), p. 58.56 See Fung/Hsieh (2002a), p. 23, and Ackermann et al. (1999), p. 864.57 Survivorship bias has been well documented in the mutual fund literature, see for example
Grinblatt/Titman (1989), Brown et al. (1992), Brown/Goetzmann (1995) and Malkiel (1995).58 This special research-driven bias is called look-ahead bias or multi-period sampling bias which is
a sub-category of survivorship bias. This bias addresses the problem of implicitly selecting funds
from a data sample, that survive a number of consecutive periods, see for example Ackermann
et al. (1999), pp. 869-870, Baquero et al. (2002), pp. 3-4, and Horst et al. (2001).59 See Fung/Hsieh (2002a), p. 24.
4 Description and Analysis of Hedge Fund Data 130
the database and the resulting consequences have to be analyzed when survivorship
bias in general is measured. The resulting biases are usually aggregated under the
so-called “termination bias”61 and the percentage of hedge funds which disappear
from a database in a given period (generally 1 year) is captured in the so-called
“attrition rate”.62
When a fund decides to stop reporting to a hedge fund database provider it usually
has reached a capacity or investor limit. The database is no longer needed as a
marketing instrument. In these exit situations we would expect the database to
underestimate the performance of the hedge fund universe as successful funds usually
quickly attract new investors and therefore reach capacity or investor limits more
rapidly than the average hedge fund.63,64 In the case of a delisting we would expect
a rather negative future performance for a fund as database providers usually delist
funds that are likely to harm their reputation for providing reliable information
to their customers.65 When funds that cease operations are no longer part of the
database,66 this would entail no special biases as the fund would leave the hedge
fund universe and the database at the same time. But there is another important
bias when these disappearing funds do not report the final periods leading up to and
including their liquidation. It is conceivable that hedge funds lose substantial value
60 For this terminology see for example Fung/Hsieh (2000), p. 294. The subset of funds that
ceased operations is often additionally termed as “dead” funds in the literature on survivorship
bias, see for example Fung/Hsieh (2002a), p. 24.61 See for example Ackermann et al. (1999), p. 864.62 See for example Bares et al. (2001), p. 7.63 See Signer (2002), p. 32.64 In some publications on database biases this subset of survivorship bias is called “self-selection
bias”, see for example Ackermann et al. (1999), p. 864. We will not use the term “self-selection
bias” here because of a possible mix-up with selection bias that is outlined in Section 4.2.1.3 on
the following page. The self-selection bias analyzed in these publications considers the problem
when hedge funds refuse to report to a database provider.65 See Fung/Hsieh (2002a), p. 24.66 According to Liang (2000) poor performance is the main reason for a fund’s disappearance from
hedge fund databases, see Liang (2000), p. 315.
4 Description and Analysis of Hedge Fund Data 131
in the period subsequent to reporting. Investors in hedge funds that cease operations
may experience discounts in value due to the liquidation of the underlying fund
holdings, and a delay in the redemption of proceeds.67 Even studies that explicitly
address the termination bias by studying the returns of discontinued hedge funds
may suffer from this liquidation bias. In order to minimize this effect database
providers usually make efforts including multiple follow-up phone calls and faxes for
about one year after disappearance.68
4.2.1.3 Selection Bias
The voluntary reporting to hedge fund databases is another source for data biases.
As hedge funds are not allowed to advertise publicly, voluntary reporting is a way
to distribute information and attract investors.69 Because of hedge fund managers
that refuse to report their results,70 the databases do depart from the actual hedge
fund universe and are not fully representative for the hedge fund population.71 The
resulting bias is called “selection” bias or “self-selection” bias.72
There are two possible reasons for this selection bias. On the one hand hedge fund
managers with a superior track record will be more likely to attract new investors
by the usage of a hedge fund database as a marketing tool. A manager with a
poor historical performance will think twice if the comparison to other hedge funds
in such a database is the right sales channel. As a consequence hedge funds in a
database would tend to have better performance than the excluded funds.73 An
67 See Ackermann et al. (1999), p. 867.68 See Ackermann et al. (1999), p. 867.69 See for example Liang (2000), p. 312.70 The restrictive inclusion methodology of database providers could be another reason for not
including certain hedge funds to a database, see for example Fung/Hsieh (2002a), p. 24. But we
will assume that the relevant hedge fund population does meet the respective inclusion criteria.71 See Fung/Hsieh (2000), p. 299.72 This problem does not exist with mutual funds, as mutual funds must publicly disclose their
results.
4 Description and Analysis of Hedge Fund Data 132
illustrative numerical example on this part of selection bias and the consequences
can be found in Lo (2001).74
On the other hand if a hedge fund performs consistently well and does not need ad-
ditional capital, the manager might also refuse to disclose the numbers to a database
provider. So rather attractive funds from the hedge fund universe could not be in-
cluded by data providers. This reason for not reporting to any database would lead
to a database performance that is biased downward as good-performing funds from
the population would not be taken into account.75
The two reasons for selection bias partially offset each other which limits the mag-
nitude of this bias. Furthermore the net effect of the selection bias is ambiguous
as the sign of this bias depends on which group of hedge funds dominates in the
population.76
4.2.1.4 Instant-History Bias
The backfilling or instant-history bias is addressing a problem on the database level.
When data providers add a new hedge fund to their database, the historical fund
performance is often backfilled into the database.77 If a database provider does
ignore this problem and backfills the complete available history of a new database
member, these funds enter the databases with so-called instant history78 as the
database will include the fund’s return history prior to the fund’s entrance date.79
Because of this instant-history bias the database samples usually overestimate the
73 See for example Ennis/Sebastian (2003), p. 104, Baquero et al. (2002), p. 4, and Fung/Hsieh
(2002a), p. 24.74 See Lo (2001), pp. 20-21.75 See Lhabitant (2002), pp. 134-135.76 See Lhabitant (2002), p. 135, Fung/Hsieh (2000), p. 299, and Fung/Hsieh (2002a), p. 25.77 See for example Edwards/Caglayan (2001b), p. 1007.78 See Park (1995) for this formulation.79 See Fung/Hsieh (2002a), p. 25.
4 Description and Analysis of Hedge Fund Data 133
performance of the hedge fund population. There are three main reasons for this
bias. At first only funds that survived the incubation period are able to enter a
database and look for new investors. During this incubation period the investor
base is usually small, and if the manager is able to earn acceptable returns on this
seed capital, the manager will try to market the fund. The hedge funds that close
down before looking for a broader investor base by reporting to a database provider
will not be included or backfilled in the historical data of hedge fund database
providers.80 This part of instant-history bias is related to the survivorship bias in
Section 4.2.1.2 on page 128.
As usually only successful funds with a satisfactory track record will elect to backfill
returns, this might be another reason why the databases overestimate the hedge
fund population return.81 This part of instant-history bias is related to the selection
bias in Section 4.2.1.3 on page 131.
A third source of instant-history bias is the adjustment of the reported and backfilled
return series by the hedge fund manager. While the actual performance numbers
are usually audited, the hedge fund manager still has the freedom to wait for a good
period of performance before requesting the database inclusion or to truncate the
performance history in order to provide only the most recent and more successful
part.82
4.2.1.5 Stale-Price Bias
The stale-price bias that is also called managed-price bias refers to problems with
hedge fund return measurement and is located on the hedge fund level. As many
hedge funds trade to at least some degree, illiquid securities, the determination of pe-
riodic net asset values might become difficult.83 When assets like illiquid exchange-
80 See Ackermann et al. (1999), p. 867, and Fung/Hsieh (2002a), p. 25.81 See Fung/Hsieh (2000), p. 298, and Ennis/Sebastian (2003), p. 104.82 See Lhabitant (2002), p. 135.
4 Description and Analysis of Hedge Fund Data 134
traded securities or OTC-securities that are difficult to price, are included in a hedge
fund portfolio but do not trade (frequently) near the end of the month, hedge fund
managers have considerable flexibility in how they mark their positions for month-
end reporting.84
Given this flexibility, it is not unreasonable to assume that hedge fund managers have
an incentive to smooth the fund performance in order to report a return series with
lower risk (by reducing the reported volatility) and a lower correlation or dependence
to other asset classes. Therefore, a hedge fund’s net asset value does not necessarily
reflect the true market value of the hedge fund portfolio constituents.85
For illiquid securities that are thinly or infrequently traded the hedge fund manager
could use a price at which the security last traded (a so-called “stale” price), estimate
a price using proprietary pricing models along with broker-dealer input or simply
chose a reasonable price.86 All of these alternatives still leave considerable flexibility
to determine net asset values with the manager. Even if the portfolio constituents
are not that illiquid, the manager is able to smooth net asset values by delayed
reporting or by requesting dealer’s prices for evaluation purposes only.87 The result
of such a smoothing of return figures is an autocorrelated return series.88 This topic
is outlined in Section 4.4.3.
4.2.1.6 Consequences of Hedge Fund Database Biases
In this section the different biases and the resulting quantitative consequences for
the measurement of hedge fund performance are brought together. Figure 4.3 gives
83 See Asness (2001), p. 19.84 See Asness et al. (2001), p. 10.85 See Schneeweis et al. (2001), p. 21, Schneeweis et al. (2002), p. 14, and Fung et al. (2004), p.
23.86 See for example New (2001), p. 3.87 See Asness (2001), p. 19.88 See Getmansky et al. (2003) and Brunner/Hafner (2005), pp. 130-132.
4 Description and Analysis of Hedge Fund Data 135
a review on the different reasons for biases and already indicates the impact of
these biases. In the following the direction and magnitude of survivorship bias and
backfilling bias are discussed. Selection bias and stale-price bias will not be analyzed
as they cannot be addressed directly in empirical research.89
�����������
���
�� ��������
����
� � �����
���
����������
���
���� �����
���
���������
���
� ���������
���
� �������
�����������
���
����� !���
�� ���� �����
� ���� �� � ���
���� �����
�����
�� ��� � ���
" ���� ��
�� � ������ �
#��������
� � �� ��
���������� ��
���� ����
��������� �����
���������� ��
�� ���� �����
� �����
���������� ��
������� � ����
" ���� ��
�� � ������ �
" ������ ��
���� ��� � ��
��$� �
" ������ ��
������ � � �
��� � ���� �
��! � ���� %
���� ������
Figure 4.3: Different reasons for biases and the resulting consequences
At first survivorship bias is considered. Different studies estimated the extent of
this bias. The methodologies to quantify survivorship bias in these studies slightly
differ. In general the survivorship bias of a sample with fund data is measured as
the difference in the performance of two portfolios: an “observable” portfolio and a
“surviving” portfolio. The observable portfolio does invest equal amounts in each
fund in the data sample or database. The portfolio is rebalanced in order to maintain
equal investments in the individual funds when a new fund is added to the portfolio
or when a fund left the data sample. The capital that is returned from these defunct
funds is reinvested in the remaining funds.90
89 See for example Fung/Hsieh (2000), p. 299, and Fung/Hsieh (2002a), pp. 25-26.90 See for example Fung/Hsieh (2000), p. 294.
4 Description and Analysis of Hedge Fund Data 136
The surviving portfolio is invested equally in all funds that are in the database at the
end of the analyzed period. Therefore, the surviving portfolio represents a return
that an investor would earn if all funds that left the database during the sample pe-
riod would have been avoided. This definition is used in the empirical work of Bares
et al. (2001), Brown et al. (1999), Capocci (2001), Edwards/Caglayan (2001b),
Fung/Hsieh (2000), Liang (2000) and Liang (2001). In the studies of Liang (2000)
and Liang (2001) the surviving portfolio and the observable portfolio is determined
every year. The surviving portfolio therefore includes the funds that exist at the
end of each year and the observable portfolio is set up with all funds that have been
in the data sample during the year.
For the surviving portfolio some researchers chose a more restrictive definition and
also excluded new funds that started reporting during the examined period. This is
done in Amin/Kat (2002a) and in Brown et al. (1999). Table 4.7 on the next page
summarizes the results of these articles and research papers on survivorship bias in
hedge fund databases.
While the first, less restrictive definition of surviving funds is in general appropri-
ate to determine the survivorship bias in the data sample, the second definition,
which does exclude new funds, is suitable for investors that are interested in his-
torical parameter estimates for which at least a number of years of data should be
available.91
According to Table 4.7 the survivorship bias overestimates hedge fund performance
by an amount of 0.60% to 3.00% in annual returns when whole hedge fund databases
(not just estimates for different strategies) are considered.92 In the mutual fund
literature this bias is also well documented and estimated with 0.5% to 1.5% per
91 See Amin/Kat (2002a), p. 9.92 A very detailed analysis of survivorship bias can be found in Amin/Kat (2002a). The authors
determine the survivorship bias in the TASS-database for different subgroups according to size,
age, co-investments, leverage and the strategies of the hedge funds in the sample, see Amin/Kat
(2002a), p. 20.
4 Description and Analysis of Hedge Fund Data 137
annum.93
Bias Time New Database
p.a. frame funds
Ackermann et al. (1999) 1.10% 1988-1995 yes MAR/HFR
Amin/Kat (2002a) 1.89% 1994-2001 no TASS
Bares et al. (2001) 1.30% 1996-1999 yes FRM
Brown et al. (1999) 0.75% 1989-1995 no US Offshore
Brown et al. (1999) 3.00% 1989-1995 yes US Offshore
Capocci (2001) 0.60% 1984-1993 yes HFR/TASS
Capocci (2001) 1.20% 1994-2000 yes HFR/TASS
Edwards/Caglayan (2001b) 0.36% 1990-1998 yes MAR (MN)
Edwards/Caglayan (2001b) 3.06% 1990-1998 yes MAR (LO)
Fung/Hsieh (2000) 3.00% 1994-1998 yes TASS
Liang (2000) 0.60% 1994-1997 yes HFR
Liang (2000) 2.24% 1994-1998 yes TASS
Liang (2001) 2.40% 1990-1999 yes TASS
Table 4.7: Survivorship bias in hedge fund databases94
In order to assess the magnitude of backfilling or instant-history bias, the method-
ology is similar to the estimation of survivorship bias. In this case the difference
93 See for example the research on mutual fund survivorship bias in Grinblatt/Titman (1989),
Brown et al. (1992), Brown/Goetzmann (1995) and Malkiel (1995).94 See Ackermann et al. (1999), pp. 864-865, Amin/Kat (2002a), p. 12, Bares et al. (2001),
p. 24, Brown et al. (1999), pp. 102-103, Capocci (2001), p. 15, Edwards/Caglayan (2001b),
p. 1007, Fung/Hsieh (2000), p. 297, Liang (2000), p. 325, and Liang (2001), pp. 15-17. In
Edwards/Caglayan (2001b) only the survivorship bias for market neutral hedge funds and long-
only equity funds is reported. See Section 4.1 for the description of the mentioned database
providers. Financial Risk Management (FRM) is an independent investment services company,
see www.frmhedge.com.
4 Description and Analysis of Hedge Fund Data 138
in the performance of an observable portfolio (as defined previously) and a second
portfolio which should not be affected by backfilling is measured. A common ap-
proach to determine the second portfolio is to eliminate the first one or two years
of reported data for every fund as these years should contain the most backfilled
data.95 The months that are dropped for the second portfolio are usually called
“incubation period”.96
Bias Time Incu- Database
p.a. frame bation
Ackermann et al. (1999) 0.10% 1988-1995 2 years MAR/HFR
Capocci (2001) 0.90% 1984-2000 1 year HFR/TASS
Capocci (2001) 1.20% 1994-2000 1 year HFR/TASS
Edwards/Caglayan (2001b) 1.17% 1990-1998 1 year MAR
Fung/Hsieh (2000) 1.40% 1994-1998 1 year TASS
Table 4.8: Instant-history bias in hedge fund databases97
Table 4.8 summarizes the results of some articles and papers on backfilling bias in
hedge fund databases. With the exception of Ackermann et al. (1999), empirical
research reports a backfilling bias of 0.90% to 1.40% percent per annum.98 As
expected, the instant-history bias overestimates the performance of the hedge fund
population.99
95 See Ackermann et al. (1999), p. 868.96 See for example Fung/Hsieh (2002a), p. 25.97 See Ackermann et al. (1999), pp. 868-869, Capocci (2001), pp. 15-16, Edwards/Caglayan
(2001b), p. 1007, and Fung/Hsieh (2000), p. 298.98 The results in Ackermann et al. (1999) are not supported by other researchers and are attributed
to the particular database used in Ackermann et al. (1999), see for example Brown et al. (2001),
p. 1874.99 With most index providers backfilling bias is not a problem as the added performance history
of new funds does not affect historical index values.
4 Description and Analysis of Hedge Fund Data 139
4.2.2 Weighting Schemes for Hedge Fund Indices
Besides the biases that hedge fund indices inherit from the corresponding hedge fund
databases, the weighting scheme of hedge fund indices might entail further problems
for the interpretation and the usage of hedge fund index data. We will discuss the
two most common weighting schemes in asset management, that is equal weighting
and value weighting.100
As Table 4.1 on page 119 indicates, most index providers offer equally weighted hedge
fund indices. The monthly or quarterly index series are calculated as arithmetic
average over the returns from funds in the respective category. For a virtual investor
in these indices this weighting scheme does imply a certain investment strategy that
is periodically rebalancing the portfolio by selling “winners” and buying “losers”
from the underlying hedge fund sample. Money that was allocated equally in these
hedge funds is shifted within the sample from well-performing to underperforming
funds. For the investor the corresponding indices therefore represent contrarian
asset allocation strategies on the hedge fund sample.101
Other index providers offer value-weighted index series, where the weights for dif-
ferent funds are directly connected to the net asset value of the hedge fund. For an
investor such a value-weighting of portfolio constituents might be a more common
and natural portfolio strategy. This weighting-scheme represents a strategy based
on momentum. “Winners” that performed well over the last period will be able to
increase their weight in the portfolio or index, while “losers” that underperformed
will accordingly reduce their weight.102
100 Two weighting schemes that are additionally mentioned in Table 4.1 on page 119 are a weighting
based on principal components analysis (PCA) and capacity weighting (cw). As both weighting
schemes do not represent interpretable portfolio strategies they will not be discussed in the
following. In the case of the PCA this implicit weighting is applied on the level of indices with
different index methodologies and is therefore not interpretable anyway.101 See Fung/Hsieh (2002a), p. 26.102 See Fung/Hsieh (2002a), p. 26.
4 Description and Analysis of Hedge Fund Data 140
Considering the different weighting schemes therefore reveals another source for bi-
ases in hedge fund indices. The differences in equal weighted and value weighted
indices are for example analyzed in Fung/Hsieh (2002a) using the composite index
from HFR which is equally weighed and the value weighted CSFB/Tremont com-
posite index.103 The differences on an annual basis sum up to an amount of for
example −9.1% in 1994 and +9.5% in 1997, respectively.104
Using equally weighted indices might be suitable for an analysis concerning the aver-
age return of an investment in a particular strategy. Value weighting is focused on a
portfolio approach and is therefore more useful in the context of portfolio considera-
tions with the individual funds in the database. Hence, for a broad hedge fund index
a value weighting scheme that implies allocations according to the capitalization (or
net asset value) might be appropriate.
4.2.3 Fund of Hedge Funds Data
For estimating the result from investing in hedge funds it might be useful and
natural to look at the results of hedge fund investors that is fund of hedge funds.105
Figure 4.4 on the next page gives an impression on the differences between composite
hedge fund indices and fund of hedge fund indices.
Data providers calculate composite and strategy indices from the hedge funds in
the database by an explicit weighting scheme, usually equal weighting. This explicit
weighting is also applied when it comes to averaging over fund of hedge funds in
order to determine an index for this group. The main difference between composite
or strategy indices and fund of hedge funds indices is the weighting scheme for the
target funds as fund of hedge funds indices implicitly apply value weighting in terms
103 See Section 4.1.2 for a description of these index providers.104 See Fung/Hsieh (2002a), p. 26.105 See Fung/Hsieh (2000), p. 300.
4 Description and Analysis of Hedge Fund Data 141
�����
����
�����
����� �������
�����������
������������������������
���������������������������������������������������
� �����������
� ��������������
�� ���������
�������
�����
����� �����
�������
! � � ���������������"
�����
����������� �����
�������
Figure 4.4: Constituents and weighting schemes of hedge fund indices
of their allocations. So the consequences for a hedge fund investor might be mirrored
more appropriately by fund of hedge fund indices than with most composite indices
for hedge fund databases.
Compared to general composite indices, fund of fund indices offer numerous ad-
vantages when it comes to measuring hedge fund performance. Figure 4.5 on the
following page summarizes the most important advantages of fund of hedge fund
indices and some of the corresponding drawbacks.
An advantage of fund of hedge fund data over data from individual funds is the
accurate and usually audited performance information on a timely basis. This is
one of the services the fund of fund management should provide to the investor.
A drawback concerning this usually timely reporting can be seen in occasional late
reporting due to the aggregation of data from the target funds.106
Another advantage of fund of hedge funds is that the biases and return differ-
106 See Fung/Hsieh (2000), p. 301, and Fung/Hsieh (2002a), p. 30.
4 Description and Analysis of Hedge Fund Data 142
��������� ����������
����������������
����� ������������ ��
��� ����������
������������� �
������ �����
����������������� ���
��������� �������������������
����� ����� ������
� ��������������������������
��������������� ���������
����� ����� ������
������������������
Figure 4.5: Advantages and disadvantages of fund of hedge fund indices
ences from the miscellaneous weighting schemes of various index providers are much
smaller when fund of hedge fund indices are considered.107
Fund of hedge fund data is less vulnerable to biases from data samples. According to
Fung/Hsieh (2000) fund of hedge fund data exhibits less than half the survivorship
bias and backfilling bias of individual hedge funds. The selection bias is also not a
real problem with fund of hedge funds as they invest in a large portfolio of target
funds and are therefore less affected by capacity constraints on the fund of hedge
funds level.108
Fund of hedge funds also avoid most data problems and biases from individual hedge
funds. As fund of hedge funds usually invest in individual funds that cease operations
survivorship bias is avoided as dead funds and funds that stopped reporting to
database providers remain in the track record of the fund. Selection bias is also
not relevant as the performance of individual funds is included in the fund of hedge
fund even if the fund manager might choose not to report to a database provider.
Instant-history on the level of individual hedge funds is not found in fund of hedge
107 See Fung/Hsieh (2002a), p. 31, and the data in Fung/Hsieh (2000), p. 302.108 See Fung/Hsieh (2000), p. 301.
4 Description and Analysis of Hedge Fund Data 143
funds data either, as the (audited) performance history is never backfilled when new
funds are added to the portfolio. Finally stale or managed price bias should be less
severe with fund of hedge funds as the monitoring is usually more sophisticated than
with traditional hedge fund investors.109
Besides these favorable characteristics of fund of hedge fund indices there are also
some disadvantages of these indices when used as proxies for hedge fund investments.
At first only a smaller fraction of the single hedge funds universe is included in fund
of hedge funds. The reason is that single hedge funds that are not reporting to fund
of hedge fund investors are not part of the measured performance. Furthermore
expenses and fees (management fees and performance fees) on the fund of hedge
funds level are slightly biasing the reported performance. Reported fund of hedge
funds performance is usually measured net of fees. This includes the subtraction of
fees on the target fund level and on the level of the fund of fund management. As
these fees on the fund of hedge fund level are not typical for a general hedge fund
investor the performance is biased slightly downward.110
The cash held by fund of fund managers for redemptions is another source of a
minor bias in fund of hedge fund index data.111 As an individual hedge fund investor
does not need this liquidity the fund of hedge fund performance as a proxy for the
performance of a hedge fund investor is biased downwards.112
109 See Fung/Hsieh (2000), p. 301, and Fung/Hsieh (2002a), pp. 30-32.110 See Fung/Hsieh (2000), p. 302.111 See Section 2.3.6 for details on redemption policies.112 See Fung/Hsieh (2000), p. 302.
4 Description and Analysis of Hedge Fund Data 144
4.3 Data Selection from the Hedge Fund Index
Universe
If we take a look at the variety of hedge fund strategies, the corresponding hedge fund
indices and the biases, the question is which index should be used as a benchmark
or as a proxy for hedge fund investments in general.
As a result from Section 4.2.2 on weighting schemes we would recommend either
equally weighted fund of hedge fund indices or composite indices with value weight-
ing. These value weighted composite indices and fund of hedge fund indices both
apply a value weighting on the single hedge fund level and therefore give a useful
proxy for the result of actual (diversified) hedge fund allocations.
Because of the problems outlined in Section 2.3.6, a market portfolio of (all observ-
able) hedge funds in a database is not an adequate passive investment alternative
because of liquidity issues like minimum investment limits. Furthermore the ar-
guments that where brought forward in Section 4.2.3 suggest the usage of fund of
hedge fund indices instead of a general composite index.
Hence we will use fund of hedge fund indices as proxies for the performance of
hedge fund investments in the following. Table 4.9 on the next page lists the index
providers that where introduced in Section 4.1. This overview gives information
if the index provider calculates a fund of hedge funds index and which weighting
scheme is employed. Additionally information on the backfilling and the verification
of hedge fund data is given.
The column “Backfilling of Data” indicates whether historical performance is back-
filled when a fund enters the database. For most indices backfilling is not a problem
as newly added hedge fund history does not affect the history of indices.113 The ver-
4 Description and Analysis of Hedge Fund Data 145
Fund of Hedge Weight- Backfilling Verification
Funds Index ing of Data of Data
AAC yes ew yes —
ABN/EH no ew yes no
Altvest/IF yes ew no yes
Barclay yes ew no yes
Bernheim no ew no yes
BlueX no ew/vw no —
CISDM yes median no yes
CSFB no vw no yes
CSFB inv no vw no no
Dow Jones no ew no yes
EACM no ew no no
EDHEC yes PCA — —
Feri no ew no yes
FTSE no cw no yes
Hennessee no ew no yes
HFI only HFI IH median once yes
HF-NET yes ew yes no
HFRI yes ew no no
HFRX no ew/vw no yes
Mondo yes ew&vw no —
MSCI no ew/vw no yes
MSCI inv no ew no yes
S&P no ew no yes
VAN no ew no no
Table 4.9: Summary on reasons for hedge fund database biases
4 Description and Analysis of Hedge Fund Data 146
ification of data by the database provider could reduce stale-price bias on the fund
of hedge funds level to some extent. But as this problem is much more severe on the
target or individual fund level when composite or strategy indices are calculated,
the impact of managed fund of hedge funds prices on the performance measurement
should be minimal.
Only eight index providers calculate fund of hedge fund indices, all of which represent
equal weighted averages of the underlying performance data. These indices are
gathered in Table 4.10 with additional information on the length of the historical
time series and whether any restrictions are imposed on the funds for the index.
Provider and Index History Restrictions
or Focus
AAC Fund of Hedge Funds Benchmark Jan 1997 - Dec 2004 —
Altvest/IF Subindex : FoF Jan 1993 - Dec 2004 —
Barclay Fund of Funds Jan 1998 - Dec 2004 —
CISDM Fund of Fund Median Mar 1997 - Dec 2004 —
EDHEC Funds of Funds Jan 1997 - Dec 2004 —
HFI InvestHedge Composite Jan 2002 - Dec 2004 Asian-Pacific
HFRI FoF Composite Jan 1990 - Dec 2004 —
Mondo Hedge Indice Generale Jan 2002 - Dec 2004 Italian
Table 4.10: Fund of hedge fund index providers
These indices should give a relatively unbiased estimate for the performance of
hedge fund investments, so they are in the focus of Section 4.4 when the statistical
properties of hedge fund indices are analyzed. As the fund of hedge fund indices
113 HEDGEFUND.NET is an exception as “historical” index values are recalculated when a fund
enters the database with instant history. Therefore, the index series is excluded from further
analysis.
4 Description and Analysis of Hedge Fund Data 147
from HFI and Mondo offer historical data for a relatively short period of time we will
exclude them from the database. Additionally the indices with the shorter return
time series are focused on fund of hedge funds from special and therefore not very
representative regions like Italy and the Asian-Pacific region. This leaves us with
six hedge fund index return series for the data analysis in Section 4.4.
4 Description and Analysis of Hedge Fund Data 148
4.4 Statistical Properties of Fund of Hedge Fund
Index Data
In this section some important statistical properties of fund of hedge fund index
returns are examined. Such a detailed analysis of the return data is essential for
the modeling of portfolio returns in Chapter 5. In this section we examine the
unconditional distributions and certain time series properties of fund of hedge fund
return data. All of these index returns are measured in US Dollar. Figure 4.6
gives an impression on the performance of the six fund of hedge fund indices under
consideration since 1997.
1997 1999 2001 2003 2005
100
150
200
250
300
AACFund of Hedge Funds Benchmark
1997 1999 2001 2003 2005
100
150
200
250
300
Altvest/IFSubindex : FoF
1997 1999 2001 2003 2005
100
150
200
250
300
BarclayFund of Funds
1997 1999 2001 2003 2005
100
150
200
250
300
CISDMFund of Funds Median
1997 1999 2001 2003 2005
100
150
200
250
300
EDHECFunds of Funds
1997 1999 2001 2003 2005
100
150
200
250
300
HFRIFoF Composite
Figure 4.6: Performance of fund of hedge funds indices since 1997114
From Figure 4.6 we can already see that the development of the historical perfor-
114 All six indices are set to a value of 100 at the beginning of 1997.
4 Description and Analysis of Hedge Fund Data 149
mance for five of the six indices is pretty similar over the period beginning in the
year 1997. Only the performance of the Altvest/IF fund of hedge fund index dif-
fers from other performance histories. The best absolute performance since 1997 is
reported for this return series.115 Since January 1997 a total net return of 180.82%
accumulated while most other fund of hedge fund indices offered absolute returns
of 80%-100% over this period of eight years.116 Table 4.11 summarizes the absolute
returns of the six hedge fund indices.
Provider Index Absolute Performance
since 1997
AAC Fund of Hedge Funds Benchmark 102.76%
Altvest/IF Subindex : FoF 180.82%
Barclay Fund of Funds 82.38%
CISDM Fund of Fund Median 78.77%
EDHEC Funds of Funds 112.03%
HFRI FoF Composite 79.60%
Table 4.11: Fund of hedge fund index performance
In the following we will take a closer look at the monthly returns of these index
series. At first in Section 4.4.1 the return definitions that are used in the following
sections are introduced. Section 5.2.2 gives more information on the unconditional
distribution of fund of hedge fund index returns and Section 4.4.3 describes certain
time series properties of these historical index returns.
115 See Section 4.1 for possible reasons for this outstanding performance.116 In the case of the Barclay Fund of Funds index we only have seven consecutive years of monthly
return data in our sample. For the CISDM Fund of Fund Median the return data is available
since march 1997.
4 Description and Analysis of Hedge Fund Data 150
4.4.1 Return Definitions
In financial modeling the most important (random) variable is the “return” of assets
or investments in general. This return can be defined in different ways. As the
understanding of the underlying return definition is crucial we will introduce the
two different return concepts of simple and log returns.
The simple return is applied in many contexts.117 Denote by Pt the price of an asset
or index at date t that is adjusted for any payments, stock splits, etc. The simple
(net) return on the asset or index, Rt,118 between dates t− 1 and t is defined as119
Rt =Pt
Pt−1
− 1. (4.1)
Sometimes the term “discrete” return as opposed to the “continuous” return that is
introduced in the following is also employed for this definition.120 The simple return
of a portfolio of assets with weights wi und returns Ri,t, i = 1, ..., N , is calculated
according to
RP,t =N∑
i=1
wiRi,t. (4.2)
This linear relationship between simple portfolio returns and the returns of the
portfolio components simplifies the calculations and is one of the reasons why this
return definition is used in the portfolio model outlined in Chapter 5.
If the compounding happens in continuous time, i.e. interest is paid every instant,
a continuously compounded return or so-called log return can be calculated.121 The
log return, rt, for period t is defined as
rt = ln
(Pt
Pt−1
)= ln (1 + Rt) . (4.3)
117 See Dorfleitner (2002), pp. 237-238 for several applications of discrete returns with financial
problems.118 (1 + Rt) describes the simple gross return.119 See for example Campell et al. (1997), p. 9.120 See for example Bamberg/Dorfleitner (2002), p. 866.121 See for example Copeland et al. (2005), pp. 889-890, for the limiting case of increasing com-
pounding periods.
4 Description and Analysis of Hedge Fund Data 151
This return definition is especially suited for continuous time models in option pric-
ing or time series analysis as it is additive over time.122 When portfolios of assets are
examined the corresponding log return of a portfolio is calculated according to123
rP,t = ln
(N∑
i=1
wieri,t
). (4.4)
In many textbooks the approximation
rP,t ≈N∑
i=1
wiri,t (4.5)
is given.124 In empirical research using the approximation from Equation 4.5 instead
of the exact result in Equation 4.4 is only a minor problem, as returns that are mea-
sured over short periods of time are usually close to zero. Therefore, the numerical
difference between the two return definitions is small, as Figure 4.7 illustrates. This
absolute return difference is displayed on the right hand plot of Figure 4.7.
-0.10 -0.05 0.0 0.05 0.10
-0.1
00.
00.
10
Simple Return
Log Return
-0.10 -0.05 0.0 0.05 0.10
0.0
0.00
5
Simple Return - Log Return
Figure 4.7: Differences between simple and log returns125
As we can see the difference between the two return definitions depends on the
absolute value of the returns under consideration. As the simple return can be
122 See Dorfleitner (2002), pp. 221-222 and p. 237.123 See for example Bamberg/Dorfleitner (2002), p. 866.124 See for example Campell et al. (1997), p. 12.125 See for example Dorfleitner (2002), p. 219, for a similar figure.
4 Description and Analysis of Hedge Fund Data 152
expressed as a function of the log return the difference can be clarified in a Taylor
series or more precisely Maclaurin expansion:126
Rt = ert − 1 =∞∑i=0
rit
i!− 1 =
∞∑i=1
rit
i!= rt +
∞∑i=2
rit
i!. (4.6)
As Equation 4.6 points out the two return definitions differ in terms of second and
higher order. This leads to the common relationship Rt ≥ rt. With returns close
to zero the resulting difference is very small.127 The size of these return differences
for simple returns is illustrated in Figure 4.7 on the previous page for the interval
[−0.10, 0.10] .
A major discrepancy between the simple and the log return is their domain. While
the support of the log return rt is the real line R, the simple return Rt can only
attain values on [−1,∞) .128 The reason for this domain of simple returns is located
in the limited liability assumption for the underlying prices Pt.
In the following the term “returns” will be used for simple returns and “log returns”
will describe continuously compounded returns. As Chapter 5 presents a modeling
approach which considers only two points in time, we will usually employ (discrete)
returns and the corresponding distributions for these returns. The standard text-
book assumption of a distribution for portfolio modeling that is defined on R is not
consistent with the support of the simple return.129 This is an approximation but
the errors due to this assumption are negligible when a relatively short period of
time which corresponds to minor variation in the returns is considered. Therefore,
the probability of simple returns less than −1 becomes very small.130
126 See Dorfleitner (2002), pp. 218-219.127 See Dorfleitner (2002), pp. 219-220.128 See for example Bamberg/Dorfleitner (2002), p. 867.129 See for example Huang/Litzenberger (1988). p. 62, and Bamberg/Dorfleitner (2002), p. 867.130 As the error is usually small we will also assume continuous distributions that are defined on
the real line R for the random variable Rt in the course of modeling portfolio distributions in
Chapter 5.
4 Description and Analysis of Hedge Fund Data 153
4.4.2 Unconditional Return Distributions
This section provides a detailed analysis of the unconditional return distributions of
fund of hedge fund indices. Here we will not consider the chronology of the different
index returns.131 At first we give an impression on the empirical return distributions
by descriptive statistics. Furthermore distributional parameters for the population
of hedge fund index returns are estimated from these historical return realizations.
Finally one of the foundations of modern portfolio theory, the assumption of nor-
mally distributed returns is checked by statistical test in Section 4.4.2.3.
4.4.2.1 Descriptive Statistics
The purpose of this section is to analyze and compare the descriptive statistics of the
indices under consideration. We present aggregate statistics for the realized monthly
returns of the index series from the six data providers AAC, Altvest/IF, Barclay,
CISDM, EDHEC, and HFRI. At first the descriptive statistics for the empirical
distribution of the discrete returns are given in Table 4.12.
No. Mean Mean SD Skew- Excess
p.a. ness Kurtosis
AAC 96 0.75% 9.36% 1.40% 0.255 3.625
Altvest/IF 144 1.29% 16.64% 2.32% 0.903 3.904
Barclay 84 0.73% 9.10% 1.48% 0.413 4.243
CISDM 94 0.63% 7.81% 1.30% -1.139 8.183
EDHEC 96 0.80% 10.05% 1.75% 0.245 3.214
HFRI 180 0.81% 10.22% 1.62% -0.255 4.151
Table 4.12: Descriptive statistics for fund of hedge fund index returns since inception
131 See Section 4.4.3 for a brief analysis of hedge fund data over time.
4 Description and Analysis of Hedge Fund Data 154
Table 4.12 reports the descriptive statistics for the indices over the whole available
index history. Therefore, the number of monthly index returns in the database
differs substantially. While HFR features index data beginning in January 1990
the index series from CISDM offers monthy returns starting in March 1997 and the
fund of hedge fund index of Barclay is calculated since January 1998. The calculated
descriptive statistics since inception in Table 4.12 show considerable differences in
the means of the various return series. Especially the mean of the Altvest/IF index
with an index history that begins in January 1993 is obviously above the mean return
of other fund of hedge funds indices. The standard deviations (SD) are relatively
homogeneous. Again, for the Altvest/IF returns the highest value for the descriptive
statistic is derived.132 In order to obtain comparable data samples with few special
items133 we will concentrate in the following on the period beginning in January
1997 which leads to sampling distributions with less observations for the HFR and
Altvest/IF index. The descriptive statistics for this shorter period are reported in
Table 4.13 on the next page.
In Table 4.13 we can see the descriptive statistics for the final sample that is used
in the further data analysis. The standard deviations and the skewness coefficients
for the two truncated index series published by Altvest/IF and HFRI are slightly
higher than before while the descriptive statistic for the kurtosis of the samples does
not change dramatically in the case of HFRI. The kurtosis coefficient for Altvest/IF
substantially decreases from 3.9 to 3.0 indicating extreme returns over the period
before 1997. When we relate the descriptive statistics standard deviation and mean
by calculating the coefficients of variation we get a relatively homogeneous picture.
132 The sample mean µ for the return data is given by µ = 1T
∑Tt=1 Rt, the sample variance σ2 by
σ2 = 1T
∑Tt=1 (Rt − µ)2 and the sample standard deviation by σ. The sample skewness SK can
be calculated according to SK = 1Tσ3
∑Tt=1 (Rt − µ)3, and the sample (excess) kurtosis KU by
KU = 1Tσ4
∑Tt=1 (Rt − µ)4 − 3, see for example Campell et al. (1997), pp. 16-17.
133 Most indices with a long history suffer biases like those outlined in Section 4.2 especially for
their early years.
4 Description and Analysis of Hedge Fund Data 155
No. Mean Mean SD Skew- Excess
p.a. ness Kurtosis
AAC 96 0.75% 9.36% 1.40% 0.255 3.625
Altvest/IF 96 1.12% 14.25% 2.68% 1.042 2.979
Barclay 84 0.73% 9.10% 1.48% 0.413 4.243
CISDM 94 0.63% 7.81% 1.30% -1.139 8.183
EDHEC 96 0.80% 10.05% 1.75% 0.245 3.214
HFRI 96 0.63% 7.81% 1.83% -0.190 4.234
Table 4.13: Descriptive statistics for fund of hedge fund index returns since 1997
The values for the coefficient of variation (based on the monthly returns) range from
1.9 to 2.9.134
The range in skewness135 and kurtosis136 values is rather broad over the six fund of
hedge fund indices. The descriptive statistics for the skewness parameter give am-
biguous information about the shape of fund of hedge funds index returns. While
the skewness is negative for the indices from CISDM and HFRI, implying more ob-
servations are found in the longer left tail of the empirical distribution, the other
index returns series are rather skewed to the positive side. The sample excess kur-
tosis for all indices indicates that the realized returns might be more peaked and
more fat-tailed than a normal distribution would imply.137 The samples of all index
134 Here coefficients of variation (CoV ) are determined based on CoV = σµ .
135 In general the skewness measures the degree of asymmetry of a distribution. This standardized
third moment is usually positive when a distribution has a longer positive than negative tail,
see for example Thode (2002), pp. 44-45.136 The kurtosis measures the peakedness of a distribution. It is based on the fourth standardized
moment and is usually reported as “excess” kurtosis which is the fourth standardized moment
minus 3. This makes the “excess” kurtosis of a normal distribution equal to zero. Higher
kurtosis usually implies that more of the variance of a distribution is due to extreme deviations
form the mean. These distributions are usually called “fat-tailed”. See for example Thode
(2002), pp. 44-45, for different fat-tailed probability densities.
4 Description and Analysis of Hedge Fund Data 156
returns are leptokurtic with an excess kurtosis greater than zero.
For the reader who is interested in the distributional characteristics of log returns
Table 4.14 gives the descriptive statistics for log returns that have been calculated
from the same fund of hedge fund index series.
No. Mean Mean SD Skew- Excess
p.a. ness Kurtosis
AAC 96 0.74% 8.84% 1.39% 0.137 3.815
Altvest/IF 96 1.08% 12.91% 2.62% 0.890 2.654
Barclay 84 0.72% 8.58% 1.47% 0.277 4.340
CISDM 94 0.62% 7.42% 1.30% -1.318 9.064
EDHEC 96 0.78% 9.39% 1.73% 0.108 3.383
HFRI 96 0.61% 7.32% 1.82% -0.364 4.639
Table 4.14: Descriptive statistics for index log returns since 1997
Obviously the unconditional distributions of fund of hedge fund index returns are
skewed like the unconditional distributions of simple returns and are also fat-tailed.
The histograms for the return series since 1997 in Figure 4.8 on the following page
provide further information on the general shape of the empirical distribution of re-
alized returns. The inspection of the plots in Figure 4.8 underscores the results for
the descriptive statistics in Table 4.13 on the previous page. The distributions are
not symmetrical (especially the unconditional return distributions of the Altvest/IF,
EDHEC and HFRI indices) and are more peaked and more fat-tailed than a nor-
mal distribution. While especially the empirical distributions of the AAC and the
CISDM indices are very peaked, all the return series exhibit more observations in
137 There are different definitions of fat-tailedness, see Bamberg/Dorfleitner (2002), pp. 868-869.
We are interested in a comparison to the normal distribution therefore we relate kurtosis values
to the kurtosis of the normal distribution.
4 Description and Analysis of Hedge Fund Data 157
-0.10 -0.05 0.0 0.05 0.10
010
2030
40
AAC-0.10 -0.05 0.0 0.05 0.10
010
2030
40Altvest/IF
-0.10 -0.05 0.0 0.05 0.10
010
2030
40
Barclay
-0.10 -0.05 0.0 0.05 0.10
010
2030
40
CISDM-0.10 -0.05 0.0 0.05 0.10
010
2030
40
EDHEC-0.10 -0.05 0.0 0.05 0.10
010
2030
40
HFRI
Figure 4.8: Histograms for fund of hedge fund index returns since 1997
the tails than we would expect for a normally distributed random variable.138
More detailed information on the shape of the sampling distribution is aggregated
in Table 4.15 on the next page and Figure 4.9. The quartiles and the 1%- and 5%-
quantiles for the fund of hedge fund index return distributions are given in Table
4.15. We can infer that the quantiles of the empirical distributions are relatively
homogeneous. Only the Altvest/IF hedge fund of funds index shows slightly different
extreme quantiles (0.01-quantile or 0.05-quantile).
The corresponding boxplots in Figure 4.9 also take a look at the general shape of the
data set and additionally give some information on outliers. These boxplots display
the first and third quartiles of the data samples139, the median of the return series140
138 See also Figure 4.10 on page 163 for a illustration of the deviations from normality.139 The quartiles are determined by the ends of the boxes.140 The median is the horizontal line in the quartiles box.
4 Description and Analysis of Hedge Fund Data 158
0.01-Q. 0.05-Q. 0.1-Q. 0.25-Q. Median 0.75-Q.
AAC -1.96% -0.85% -0.63% -0.05% 0.54% 1.41%
Altvest/IF -5.11% -2.47% -1.30% -0.27% 0.67% 1.94%
Barclay -2.59% -0.88% -0.66% -0.07% 0.62% 1.41%
CISDM -2.40% -0.92% -0.49% -0.03% 0.62% 1.16%
EDHEC -2.86% -1.26% -0.85% -0.24% 0.68% 1.47%
HFRI -3.58% -1.58% -1.08% -0.35% 0.66% 1.46%
Table 4.15: Quantiles for fund of hedge fund index returns since 1997
and extreme outliers. These outliers are not within 1.5-times the interquartile range
(between the 0.25 and 0.75 quantile) from the end of the boxes. For each of the
six indices a number of 4 to 6 (positive and negative) outliers is determined. The
empirical return distribution for the Altvest/IF, EDHEC and HFRI indices show
slightly more extreme outliers than the other three data samples. As Table 4.15 and
the boxplots in Figure 4.9 report, the median values for all the fund of hedge fund
indices are located in a relatively small range from 0.54% to 0.67%.
Table 4.16 on the following page provides further information on the series of fund
of hedge fund index returns. Here the extreme values (minima and maxima) for
the time series with the corresponding months are gathered.141 The best and worst
months show certain time patterns for all of the fund of hedge fund indices. The
indices had their best months in December 1999 or February 2000 while the worst
month was (as expected) August 1998. In these months most of the different avail-
able strategy indices and all of the available composite indices with a longer per-
formance history also reported their highest and lowest returns. The minima and
maxima in the fund of hedge fund indices are therefore a logical consequence.
141 These extreme values are also displayed graphically in the boxplots of Figure 4.9 on the next
page as the highest and lowest outliers.
4 Description and Analysis of Hedge Fund Data 159
-0.0
50.
00.
05
AA
C
-0.0
50.
00.
05
Altv
est/I
F
-0.0
50.
00.
05
Bar
clay
-0.0
50.
00.
05
CIS
DM
-0.0
50.
00.
05
ED
HE
C
-0.0
50.
00.
05
HFR
I
Figure 4.9: Boxplots for fund of hedge fund index returns since 1997
Minimum Month Maximum Month
AAC -5.05% August 1998 5.27% December 1999
Altvest/IF -5.92% August 1998 11.54% February 2000
Barclay -5.10% August 1998 6.05% December 1999
CISDM -6.40% August 1998 4.24% February 2000
EDHEC -6.16% August 1998 6.66% February 2000
HFRI -7.47% August 1998 6.85% December 1999
Table 4.16: Maxima and Minima for fund of hedge fund index returns since inception
4 Description and Analysis of Hedge Fund Data 160
4.4.2.2 Parameter Estimates
To obtain more information on the univariate distributions for the fund of hedge fund
index returns the estimates of the moments of the corresponding return populations
are calculated and analyzed in this section. Parameter estimates for the different
index data samples since 1997 are reported in Table 4.17 on the following page. The
estimate for the population mean (or expected value of the distribution) µ is not
reported in the table as it equals the sample mean given in Table 4.13 on page 155.142
All of the unbiased estimates for the population mean in Table 4.13 are significantly
different from zero for reasonable levels of significance.
An estimate for the standard deviation σ of a return population is σ which is cal-
culated from a sample of returns according to143
σ = T
√1
T−1
T∑t=1
(Rt − µ)2. (4.9)
The equation for the sample skewness also gives a biased estimate of the population
skewness. An estimate of the skewness of returns is
SK = T(T−1)(T−2)
T∑t=1
(Rt−µ
σ
)3
. (4.10)
Finally an estimate of the population kurtosis is
KU = T (T+1)(T−1)(T−2)(T−3)
T∑t=1
(Rt−µ
σ
)4
− 3 (T−1)2
(T−2)(T−3). (4.11)
142 Statistically this µ is an unbiased estimate while from financial mathematics it is well known
that
RT = T
√T∏
t=1(1 + Rt)− 1 (4.7)
gives the correct (geometric) average return for the time series. But as a result of the Jensen
inequality
E[RT
]< T
√E[
T∏t=1
(1 + Rt)]− 1 = E (1 + Rt − 1) = E (Rt) (4.8)
RT underestimates the expected value of the return population, see for example Dorfleitner
(2002), pp. 223-224.143 See for example Bamberg/Baur (2002), p. 140.
4 Description and Analysis of Hedge Fund Data 161
The parameter estimates calculated from the index return series are gathered in
Table 4.17. In addition the table shows the standard errors (SE)144 and the p-
values145 for the estimated skewness and kurtosis parameters.
σ SK KU
Est. Est. SE p-value Est. SE p-value
AAC 1.41% 0.259 0.246 29.2% 3.887 0.488 0.0%
Altvest/IF 2.70% 1.059 0.246 0.0% 3.205 0.488 0.0%
Barclay 1.49% 0.420 0.263 11.0% 4.581 0.520 0.0%
CISDM 1.31% -1.158 0.249 0.0% 8.702 0.493 0.0%
EDHEC 1.76% 0.249 0.246 31.2% 3.453 0.488 0.0%
HFRI 1.84% -0.193 0.246 43.4% 4.528 0.488 0.0%
Table 4.17: Parameter estimates for fund of hedge fund index returns since 1997
As the p-values in Table 4.17 indicate, not all of the resulting estimates are different
from zero for reasonable levels of significance with a two-sided hypothesis test. When
the skewness estimates SK are considered only the fund of hedge fund index returns
from Altvest/IF and CISDM deliver skewness estimates that are significantly differ-
ent from zero for common failure rates of 1% or 5%. In the case of the Altvest/IF
fund of hedge fund index we would expect a longer positive tail of the distribution
while the results from the CISDM data leads to an asymmetric distribution with a
144 The standard error SESK
for the estimated skewness parameter SK is calculated according to
SESK
=√
6T (T−1)(T−2)(T+1)(T+3)
(4.12)
and the standard error SEKU
for the estimate of the population kurtosis KU is given by
SEKU
=√
4(T 2−1)SESK
(T−3)(T+5) . (4.13)
See for example Thode (2002) for sampling moments of the third and fourth moment.145 The p-values in Table 4.17 are calculated under the assumption of standard normally distributed
test statistics.
4 Description and Analysis of Hedge Fund Data 162
tail that extends towards more negative index returns. With all other estimates for
the skewness parameter in 4.17 the hypotheses of a symmetrical distribution cannot
be rejected because of too high standard errors for the corresponding estimates.
When we take a look at the kurtosis estimates the picture is slightly different. All
monthly index return samples lead to (excess) kurtosis estimates that deviate signif-
icantly from zero. Therefore, return distributions with these estimated parameters
are fat-tailed in comparison to a normal distribution. Here we expect more peaked
distributions and additionally more extreme values than with common normal dis-
tributions.
4.4.2.3 Tests for Normality
The results from Section 4.4.2.2 concern the higher distributional moments that are
significantly different from zero lead to an analysis of the index return distributions
as a whole. In the following we will take a closer look at the common assumption of
normally distributed returns that is one way to justify classical portfolio theory.146
A simple way to compare the sample distribution to the normal distribution is a
quantile-quantile-plot. An inspection of the quantile-quantile-plots in Figure 4.10
on the following page already indicates a potential non-normality in the index return
series. The spread in the extreme quantiles of the empirical distributions of all the
fund of hedge fund indices and the quantiles of a standard normal distribution is
indicative of a long-tailed distribution.
After the visual inspection of the plots in Figure 4.10 on the next page we would
already doubt the assumption that the analyzed index returns where drawn from
(different) normal distributions. In order to obtain statistically significant state-
ments on the distribution of returns we will perform certain statistical tests for
normality of the fund of hedge fund index data. Table 4.18 on page 164 reports
146 See Chapter 5 for more details concerning this assumption.
4 Description and Analysis of Hedge Fund Data 163
p-values on seven different normality tests: the Anderson-Darling (AD) test, the χ2
test, the Cramer-von Mises (CvM) test, the Jarque-Bera (JB) test, the Kolmogorov-
Smirnov test with Lilliefors correction (LKS), the Shapiro-Francia (SF) test and the
Shapiro-Wilk (SW) test.147
Quantiles of Standardnormal
-2 -1 0 1 2
-0.0
50.
00.
05
AAC
Quantiles of Standardnormal
-2 -1 0 1 2
-0.0
50.
00.
05
Altvest/IF
Quantiles of Standardnormal
-2 -1 0 1 2
-0.0
50.
00.
05
Barclay
Quantiles of Standardnormal
-2 -1 0 1 2
-0.0
50.
00.
05
CISDM
Quantiles of Standardnormal
-2 -1 0 1 2
-0.0
50.
00.
05
EDHEC
Quantiles of Standardnormal
-2 -1 0 1 2
-0.0
50.
00.
05
HFRI
Figure 4.10: Quantile-quantile plots for fund of hedge fund index returns
The perhaps most common procedure to test whether a univariate sample is drawn
from a normal distribution is the Jarque-Bera test.148 This test consistently refuses
the null hypothesis that the six index return samples are normally distributed for
all reasonable failure rates.149
147 For a comprehensive overview on univariate tests for normality see Thode (2002). Gnanadesikan
(1997) also introduces different methods for assessing univariate normality, see Gnanadesikan
(1997), pp. 187-194.148 See for example Amenc/Malaise/Martellini/Sfeir (2003), Amin/Kat (2003), Berenyi (2002),
Brooks/Kat (2001), and Favre/Galeano (2002) for the usage of the Jarque-Bera test in the
context of testing hedge fund returns.149 See Bai/Ng (2005) for a discussion of problems with the Jarque-Bera test when time series are
serially correlated.
4 Description and Analysis of Hedge Fund Data 164
AD χ2 CvM JB LKS SF SW
AAC 0.01% 0.00% 0.04% 0.00% 1.30% 0.00% 0.00%
Altvest/IF 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00%
Barclay 0.00% 0.00% 0.02% 0.00% 0.53% 0.00% 0.00%
CISDM 0.00% 0.00% 0.01% 0.00% 0.23% 0.00% 0.00%
EDHEC 0.04% 0.00% 0.10% 0.00% 0.71% 0.01% 0.01%
HFRI 0.05% 0.00% 0.17% 0.00% 1.76% 0.00% 0.01%
Table 4.18: P-values of different tests for normally distributed index returns
Other especially useful test for our purposes are the Anderson-Darling test and the
Shapiro-Wilk test as they give more weight to extreme deviations.150 The Shapiro-
Wilk test delivers the same results as the Jarque-Bera test with only slightly higher
p-values in the case of the EDHEC and HFRI index return time series and the
Anderson-Darling test also refuses all of the six null hypotheses of normality for
reasonable levels of significance.
With most other tests the hypothesis of normally distributed fund of hedge fund
index returns can also be rejected. At least six of the tests in Table 4.18 always
refuse the null hypothesis of a normal distribution when failure rates of less than
1% are accepted. For the index return series of Altvest/IF, Barclay, CISDM and
EDHEC all seven tests (even the Kolmogorov-Smirnov test with Lilliefors correction
which gives more weight to “deviations” in the midrange) were able to reject the
assumption of normally distributed returns for common levels of significance.151
150 See Thode (2002), pp. 143-152 for a discussion of the power of tests for univariate normality.151 For the AAC and HFRI index returns the Kolmogorov-Smirnov test with Lilliefors correction
is not able to reject the null hypothesis on a failure level of 1%. This test is especially sensitive
to deviations in the midrange and does not put much weight on extreme values. As the smaller
deviations are usually not problematic this test should be avoided for evaluation of normality,
see Thode (2002), p. 152.
4 Description and Analysis of Hedge Fund Data 165
We additionally take a look at the distributional characteristics of log returns as most
publications on hedge funds analyze wether the unconditional empirical distribution
of log returns is normally distributed.152 Table 4.19 reports the p-values for the six
different statistical tests.
AD χ2 CvM JB LKS SF SW
AAC 0.01% 0.00% 0.05% 0.00% 1.75% 0.00% 0.00%
Altvest/IF 0.00% 0.00% 0.00% 0.00% 0.01% 0.01% 0.00%
Barclay 0.00% 0.00% 0.03% 0.00% 0.73% 0.00% 0.00%
CISDM 0.00% 0.00% 0.01% 0.00% 0.16% 0.00% 0.00%
EDHEC 0.05% 0.00% 0.14% 0.00% 1.03% 0.01% 0.01%
HFRI 0.05% 0.00% 0.18% 0.00% 2.13% 0.00% 0.00%
Table 4.19: P-values of different tests for normally distributed index log returns
The statistical tests yield slightly higher p-values for all the testing procedures in-
dicating that the empirical distributions of log returns are a bit closer to normality
than the unconditional distributions of simple returns. But as the difference between
simple and log returns for the monthly data is rather small153 the statistical normal-
ity tests for log returns also reject all of the normality hypotheses on common levels
of significance. Again the highest p-values are reported for the Kolmogorov-Smirnov
test with Lilliefors correction, but for a reasonable significance level of 97.5% even
this test rejects the hypothesis of normally distributed log returns.
With the results of the statistical tests to the hypothesis of normally distributed
returns we lost one possible justification for classical portfolio theory. The second
justification involves rather strong assumptions on the preference structure of in-
152 However this is not essential in our context as we are interested in a model approach over one
period.153 See Section 4.4.1 for the differences between the return definitions.
4 Description and Analysis of Hedge Fund Data 166
vestors.154 Therefore, these results lead us to the alternative approach to portfolio
modeling that is outlined in Chapter 5.
4.4.3 Time Series Analysis
The time series behavior of fund of hedge fund index returns will not be carried
out too extensively, as the modeling approach outlined in Chapter 5 is a classical
one-period approach. Therefore, we will not go into detail at this point. However
we will take a closer look at certain aspects of the index return time series like
autocorrelation and a related unsmoothing technique for return series. But at first
the stationarity of the return series will be examined as it is a prerequisite for the
following analysis.
4.4.3.1 Stationarity
Stationarity of time series is especially important when time series models are con-
sidered. As we develop a modeling approach that delivers a one-period result, time
series modeling is not in our focus. However the analysis of stationarity is of some
relevance as a requirement for the research on autocorrelations in Section 4.4.3.2.
Another reason for stationarity analysis on hedge fund data is the use of return se-
ries in regression models. In order to determine an appropriate modeling approach,
stationarity analysis is essential.
When stationarity of a stochastic process is considered the analyzed time series has
to satisfy certain requirements. For the analysis of autocorrelation so-called weak
stationarity or covariance stationarity is a necessary assumption.155 A stochastic
process Rt is called weakly stationary or covariance stationary when it satisfies three
154 See Section 5.2.1 for more details on assumptions concerning investor preferences.155 Strong stationarity would require that the joint distribution of all sets of observations is invariant
to when the observations are made, see Greene (2000), p. 752.
4 Description and Analysis of Hedge Fund Data 167
requirements:156 E [Rt] is independent of t, V ar [Rt] is a finite, positive constant,
independent of t, and Cov [Rt, Rs] is a finite function of t − s, but not of t or s.157
So a stationary time series has time invariant first and second moments.
With test procedures we try to identify trends in the time series under consider-
ation. Deterministic time trends do not make much sense when return series are
analyzed and modeled. Therefore, we will concentrate on the identification of pos-
sible stochastic trends in the different time series. In order to test stationarity in
our data set of fund of hedge fund index returns we first perform an (augmented)
Dickey-Fuller test. This test tries to identify unit roots in time series data. There-
fore, a Dickey-Fuller test offers the null hypothesis of a unit root in the time series
under consideration which means non-stationarity against the alternative hypothesis
of stationarity of the time series. Furthermore Table 4.20 also presents the results
of the Philips-Perron unit root test. Similar to the Augmented Dickey-Fuller test
the Philips-Perron test starts with the null hypothesis that there is a unit root.
To another class of stationarity test the Kwiatkowski, Phillips, Schmidt, Shin test
(KPSS) belongs to. Here the null hypothesis is that the time series (or the process
underlying the time series) is stationary which means that no unit root exists. Table
4.20 summarizes the results for these three statistical testing procedures.
The two unit root tests (Augmented Dickey-Fuller and Philips-Perron) do reject the
null hypothesis of unit roots in the different time series at any reasonable significance
level.158 This holds true for all six indices under consideration. The stationarity
test of Kwiatkowski, Phillips, Schmidt, Shin does not reject its null hypothesis of
156 See for example Greene (2000), p. 752, Hamilton (1994), pp. 45-46, or Schlittgen/Streitberg
(1999), p. 100.157 The covariance Cov of two random variables is Cov [Rt, Rs] = E [RtRs]− E [Rt] E [Rs] and the
variance V ar (or σ2) of a random variable is defined as V ar [Rt] = E[R2
t
]− E [Rt]
2.158 Here the Durbin-Watson test statistic analyzes the autocorrelation in the residuals of the unit
root tests. Values close to 2 are reported for all the unit root tests. Therefore, the null hypothesis
of no autocorrelation in the regression residuals can not be rejected, see Section 4.4.3.2 for more
details. As a consequence the unit root tests are valid.
4 Description and Analysis of Hedge Fund Data 168
Augmented Durbin- Philips- KPSS
Dickey-Fuller Watson Perron
p-value test statistic p-value p-value
AAC 0.00% 2.02 0.00% > 10%
Altvest/IF 0.00% 2.02 0.00% > 10%
Barclay 0.00% 2.00 0.00% > 10%
CISDM 0.00% 2.01 0.00% > 10%
EDHEC 0.00% 1.99 0.00% > 10%
HFRI 0.00% 1.99 0.00% > 10%
Table 4.20: P-values of different tests for unit roots or stationarity of index returns
stationarity for permitted failure rates below 10%. None of the time series of hedge
fund index returns comes close to this or any other reasonable level of significance.
Therefore, the unit root and stationarity tests we performed would leave us with
the result that all the time series of index returns are stationary.
4.4.3.2 Autocorrelation
There is a vast amount of literature that detects autocorrelation in the returns of
hedge funds or hedge fund indices.159 As already mentioned in Section 4.2.1.5 the
reason for such autocorrelated returns can be at least partially found in the problem
of so-called stale prices for portfolio assets and the incentives of the hedge fund
management to smooth performance numbers.160
For a covariance stationary time series of returns Rt the τ -th order autocovariance
159 See for example Alexander/Dimitriu (2004), Brown et al. (2004), Conner (2003), Davies et al.
(2004), Ennis/Sebastian (2003), Fung et al. (2004), Liang (2003), Lo (2002), Okunev/White
(2002), De Souza/Gokcan (2004), Till (2003), and also the literature overview in Getmansky
et al. (2003).160 See for example Lo (2001), pp. 28-29, and Section 4.4.3.3 for more details.
4 Description and Analysis of Hedge Fund Data 169
function γ (τ) and autocorrelation function ρ (τ) are defined as:161
γ (τ) = Cov [Rt, Rt+τ ] (4.14)
ρ (τ) =γ (τ)
γ (0)=
Cov [Rt, Rt+τ ]
V ar [Rt](4.15)
The required stationarity of all the time series under consideration is analyzed and
approved in Section 4.4.3.1. We can get a first impression on the time series behavior
considering autocorrelation with an autocorrelation plot. With this tool we can
visually check the randomness in our return time series. If the returns are completely
random we would expect autocorrelations near zero for any time lag τ .162 Figure
4.11 displays the autocorrelation plots for the six fund of hedge fund return series
with a maximum time lag of 20. From the inspection of the autocorrelation plots
in Figure 4.11 we would expect that our time series are not completely random but
rather have some degree of autocorrelation between adjacent return observations.
In order to statistically test for autocorrelation in fund of hedge funds index returns
we make use of the Durbin-Watson test statistic, the Box-Pierce, and the Ljung-
Box test. The null hypothesis for these tests is that the time series exhibit no
autocorrelation. The Durbin-Watson test statistic should be located around 2 for a
time series with no autocorrelation (for lag 1). It ranges from 0 to 4. Values greater
than 2 suggest negative serial correlation, and values less than 2 suggest positive
serial correlation.163 The Box-Pierce and the Ljung-Box test try to reject the null
hypothesis that all the first τ autocorrelation coefficients are equal to zero. The
Ljung-Box test should be better suited for our needs as the test statistic is adjusted
for finite samples.164 Table 4.21 provides a summary on the autocorrelations in the
index return series (for orders 1-3) and on the results of the statistical tests for first
order autocorrelation.
161 See for example Campell et al. (1997), p. 45.162 Certainly, for a time lag of zero the resulting autocorrelation should always be equal to one.163 See for example Greene (2000), p. 538-540, or Schlittgen/Streitberg (1999), p. 19.164 See for example Zivot/Wang (2003), p. 62.
4 Description and Analysis of Hedge Fund Data 170
Lag
AC
F
0 5 10 15 20
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
AAC
Lag
AC
F
0 5 10 15 20
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
Altvest/IF
Lag
AC
F
0 5 10 15 20
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
Barclay
Lag
AC
F
0 5 10 15 20
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
CISDM
Lag
AC
F
0 5 10 15 20
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
EDHEC
LagA
CF
0 5 10 15 20-0
.20.
00.
20.
40.
60.
81.
0
HFRI
Figure 4.11: Autocorrelation plots for fund of hedge fund index returns
The statistical test support the results from the visual inspection of the autocorre-
lation plots. Considering the Ljung-Box test we would reject the null hypothesis of
no first order autocorrelation in the time series for the index return series of AAC,
Altvest/IF, Barclay and the HFRI at failure rates below only 5%. For the Barclay
and the HFRI index series the results are even significant at significance levels of
98.88% and 99.80%, respectively. Only for the CISDM time series the reported
p-value does not allow to reject the null hypothesis at reasonable failure rates.
Therefore, we can conclude that serial correlation is inherent in (most) hedge fund
return time series. In order to get an indication of the consequences of performance
smoothing165 the next section will take a closer look at the characteristics of “un-
smoothed” returns series that exhibit no significant autocorrelation.
165 Or more generally speaking to get an indication of the consequences of illiquidity of hedge fund
positions.
4 Description and Analysis of Hedge Fund Data 171
Coefficients Durbin- Box-Pierce Ljung-Box
ρ(1) ρ(2) ρ(3) Watson p-value p-value
AAC 0.30 0.15 0.03 1.37 7.93% 2.90%
Altvest/IF 0.25 0.11 0.01 1.50 9.98% 4.40%
Barclay 0.37 0.15 0.03 1.25 3.95% 1.12%
CISDM 0.23 0.10 -0.04 1.51 33.43% 20.32%
EDHEC 0.28 0.10 -0.02 1.42 13.33% 5.90%
HFRI 0.34 0.13 -0.02 1.30 0.94% 0.20%
Table 4.21: Autocorrelation in index return time series
4.4.3.3 Unsmoothing of Hedge Fund Data
As already mentioned in Sections 4.2.1.5 and 4.4.3.2 the smoothing of net asset
values is a very common problem with hedge fund data. Especially hedge fund
managers that invest in relatively illiquid securities for which there is no recent or
observable market price available have a high degree of freedom when it comes to
reporting net asset values. To report a return or net asset value figure such funds
usually use the last available traded prices or estimates of current market price for
the valuation of illiquid instruments.166 Therefore, a hedge fund manager that trades
rather illiquid securities167 has considerable discretion when marking the portfolio’s
value. Regarding the compensation and the fund’s performance statistics a hedge
fund manager is likely to smooth returns by marking the portfolio below the actual
value in a period of large positive returns in order to create some scope for periods
with lower returns.168
166 See for example Kat (2001), p. 5, and Amenc/Curtis/Martellini (2003), p. 9.167 Illiquid underlying securities can for example be found in convertible arbitrage (see Section
3.2.2) or mortgage-backed securities (see Section 3.2.4) strategies.168 See Lo (2001), pp. 28-29.
4 Description and Analysis of Hedge Fund Data 172
We assume that it is this smoothing of return figures that results in significant
first order autocorrelation in the time series of hedge fund and hedge fund index
returns.169 Therefore, we will make use of an “unsmoothing” technique outlined in
Geltner (1991), Geltner (1993), Brooks/Kat (2001) and Davies et al. (2004),170 in
order to obtain an unsmoothed, “real” return series for the six fund of hedge fund
indices.
The smoothed observable value of a fund171 at time t can be expressed as a weighted
average of the unsmoothed true value and the smoothed fund value at time t− 1. If
a single exponential smoothing approach is assumed this results in an unsmoothed
return series with returns Rt that are based on the smoothed returns R∗t according
to172
Rt =R∗
t − λR∗t−1
1− λ. (4.16)
When the coefficient λ is set to the first order autocorrelation coefficient of the
smoothed time series the unsmoothed time series will exhibit zero first order auto-
correlation. However the mean of the time series will remain the same and as the
term “unsmoothing” already implies the newly constructed return series has a higher
standard deviation than the smoothed version.173 Table 4.22 gives a summary on
the descriptive statistics of the unsmoothed return series.
As we already anticipated, the standard deviation of the unsmoothed return series
is significantly above the standard deviation of the original time series while the
169 Getmansky et al. (2003) identify the illiquidity exposure and the corresponding smoothing
outlined above as most likely explanation of the serial correlation in hedge fund returns. Further
influences that are examined in Getmansky et al. (2003) are for example market inefficiencies,
time-varying expected returns and time-varying leverage.170 Amenc/Curtis/Martellini (2003) tackle this problem in a regression approach with the use of
lagged returns, see Amenc/Curtis/Martellini (2003), pp. 9-11. Further unsmoothing techniques
are outlined in Conner (2003) and Okunev/White (2002).171 The observable value of an index is simply the average of many different funds with smoothed
observable values.172 See Brooks/Kat (2001), p. 14.173 See Davies et al. (2004), p. 21.
4 Description and Analysis of Hedge Fund Data 173
Coefficient Mean Mean SD Skew- Excess
ρ(1) p.a. ness Kurtosis
AAC -0.02 0.72% 8.98% 1.89% 0.078 3.260
Altvest/IF -0.01 1.09% 13.93% 3.47% 0.931 2.309
Barclay 0.00 0.75% 9.38% 2.20% 0.027 3.510
CISDM -0.01 0.65% 8.11% 1.65% -1.303 9.013
EDHEC 0.00 0.77% 9.63% 2.31% 0.114 2.930
HFRI -0.01 0.59% 7.27% 2.57% -0.437 4.122
Table 4.22: Descriptive statistics for unsmoothed hedge fund index returns
mean return in general remains unchanged. Therefore, the parameter estimates in
Table 4.23 also differ from the estimates in Table 4.17 for the raw returns.
σ SK KU
Est. Est. SE p-value Est. SE p-value
AAC 1.90% 0.079 0.247 74.9% 3.504 0.490 0.0%
Altvest/IF 3.49% 0.946 0.247 0.0% 2.501 0.490 0.0%
Barclay 2.21% 0.027 0.264 91.8% 3.807 0.523 0.0%
CISDM 1.66% -1.324 0.250 0.0% 9.585 0.495 0.0%
EDHEC 2.32% 0.116 0.247 64.0% 3.156 0.490 0.0%
HFRI 2.59% -0.444 0.247 7.3% 4.414 0.490 0.0%
Table 4.23: Parameter estimates for unsmoothed fund of hedge fund index returns
The estimated standard deviation exceeds the values in Table 4.17 on page 161 by
25 to 40 percent of the smoothed values. The estimates for the higher moments in
Table 4.23 do not give such a uniform picture. Some skewness and (excess) kurtosis
estimates for the unsmoothed return series are above and some below the smoothed
4 Description and Analysis of Hedge Fund Data 174
values. But in general the significance of the results stays the same as before.
4.4.4 Results
In this section we took a closer look at the statistical properties of representative
hedge fund data. Therefore, we analyzed the returns of six fund of hedge fund in-
dices: AAC (Fund of Hedge Funds Benchmark), Altvest/IF (Subindex : FoF), Bar-
clay (Fund of Funds), CISDM (Fund of Funds Median), EDHEC (Funds of Funds)
and HFRI (FoF Composite). In order to obtain comparable data samples for these
indices we concentrated on index data beginning in January 1997. The results in
this section are focused on simple returns but to complete the picture the statistical
properties of log returns were also analyzed. As the difference between simple and
log returns for the monthly data is rather small the statistical test delivered the
same results as for the simple returns.
For this historical track record we first researched into the unconditional return
distributions. Here the descriptive statistics for the sample and parameter estimates
were reported and compared. Then statistical tests for the unconditional simple
returns and the unconditional log returns were carried out. The null hypothesis of
normally distributed returns was clearly rejected for all six index series with simple
and log returns.
Examining time series properties of the fund of hedge fund index returns discloses
significant autocorrelation in the data. Additionally the analysis of stationarity
in Section 4.4.3.1 suggests that no trend component is located in fund of hedge
fund return series. As a consequence of the significant autocorrelation in the return
series a simple unsmoothing approach that is based on the assumption of exponential
smoothing of the performance by fund managers was carried out. This yielded a
substantially higher standard deviation of returns than we would expect from the
original return data and slightly adjusted higher moments.
4 Description and Analysis of Hedge Fund Data 175
4.5 Summary
This chapter lays the foundations for the modeling approach described in Chapter
5. At first we analyzed hedge fund data providers, the available hedge fund in-
dices, and potential shortcomings of or problems with hedge fund indices. Based
on this background Section 4.2 outlined the problem of biases in hedge fund data.
Survivorship bias, selection bias, instant-history bias, stale-price bias and weighting
bias were introduced and the consequences of these data problems were quantified
if possible. As a result of this analysis fund of hedge fund indices were chosen as
proxy for hedge fund investments in general. In Section 4.4 we finally worked out the
most important properties of fund of hedge fund index returns based on six major
indices. We were able to identify representative ranges for parameter estimates of
fund of hedge fund index returns. These characteristics are the rationale for the
portfolio modeling approach with hedge fund allocations that is introduced in the
next chapter.
5 Portfolio-Selection including
Hedge Funds
Over the last years hedge funds gained considerable attention from a wide range of
investors. While the focus in the process of selecting individual target hedge funds
should be on the qualitative side, quantitative considerations will lead to the actual
construction of portfolios with hedge funds in the sense of determining optimal
asset weights. In order to attain a favorable or optimal asset allocation, fund of
fund managers as well as individual hedge fund investors have to bear their whole
portfolio and the corresponding dependencies of returns for different assets or asset
classes in mind. A favorable asset allocation should usually include investments with
different dependence profiles in order to diversify risk.
The previous chapters pointed out some of the interesting special features of hedge
funds or alternative assets in general. These trading strategies usually lead to very
different risk profiles1 and especially the analysis of fund of hedge fund returns in
Section 4.4 revealed informative details on the corresponding statistical properties.
The resulting univariate and multivariate return distributions are significantly dif-
ferent from the classical model assumptions in a portfolio framework according to
Markowitz.2,3
1 See Section 3 on hedge fund strategies.2 See the original work of Markowitz (1952).3 See the analysis of hedge fund data in Section 4.
176
5 Portfolio-Selection including Hedge Funds 177
In this chapter an alternative approach for portfolio construction is presented which
is able to integrate some important characteristics of hedge fund return data. Like
Markowitz we will concentrate on the static portfolio construction problem with
regard to two points in time without interim portfolio revisions. At first we will
briefly illustrate the classical theory of portfolio selection. Then distributional char-
acteristics of return data and preference considerations for investors are presented
in Section 5.2 of this chapter. Here we will analyze the different justifications of
the traditional portfolio selection methodology. Due to problems with the classical
assumptions underlying portfolio selection, an alternative approach to the modeling
of multivariate return distributions is outlined in Section 5.3. An empirical analysis
of the introduced methodology is finally presented in Section 5.4.
5.1 Traditional Portfolio Selection
The classical portfolio selection framework of Markowitz (1952) regards two points
in time. The beliefs on the return distributions of different securities over this
single period are the starting point for this approach to portfolio selection. In
traditional portfolio theory only the first two moments (mean and variance) of each
asset’s return distribution together with coefficients of correlation are deployed to
the portfolio selection problem. With these measures of risk and return it is possible
to identify the efficient set among all asset combinations. These risk-return efficient
portfolios reward a certain amount of risk, which is statistically measured as variance
or standard deviation of portfolio returns, with the highest possible expected return.
When utility functions, which imply a certain amount of utility to risk-return trade-
offs, are evaluated for these efficient sets, it is possible to determine an optimal
portfolio allocation in terms of a maximum of expected utility.4 Then such a maxi-
5 Portfolio-Selection including Hedge Funds 178
mum utility portfolio is the optimal investment strategy over the next period for all
investors with the same expectations about the first two moments of the relevant
return distributions which decide in accordance to the specified utility function.
In order to formalize this approach let vector w represent the fractions of wealth
invested in the available securities. Here as usual the weights should sum for one.
The portfolio selection problem of determining the optimal allocations can then be
stated as follows5
maxw
E [U (V )] (5.1)
with U(.) as utility function that depends on the end of period wealth V = V0 ·
(1 + RP ), which is a function of the (simple) return RP over the period under
consideration. V0 is the initial wealth that is usually set to unity. Here the random
portfolio returns RP are obtained from RP = wTR, with R as vector of the random
returns for the available securities.
Taking expectations of the utility after expanding U(.) as a Taylor series around the
expected end of period wealth, and assuming that the Taylor series converges gives6
E [U (V )] = U (E [V ]) + 12
∂2U(V )∂V 2 E [V ] E [V − E [V ]]2 +
+∞∑i=3
1i!
∂iU(V )∂V i E [V ] E [V − E [V ]]i.
(5.2)
There are several ways to justify this traditional modeling approach (see Figure 5.1
on the next page). A first reasoning is located in the choice of the utility function U(.)
which is used to identify the optimal portfolio allocation that maximizes expected
utility. For arbitrary return distributions the choice of a quadratic utility function
leads to the mean-variance model as third and higher order derivatives in Equation
4 See Huang/Litzenberger (1988) for a detailed discussion of utility theory in the portfolio context.
Wilmott (2000a) for example gives a very brief introduction on the most relevant concepts of
utility theory, see Wilmott (2000a), pp. 477-483. For details on the Bernoulli principle, see
Bamberg/Coenenberg (2004), pp. 81-89.5 See Samuelson (1970), p. 538.6 See Huang/Litzenberger (1988), p. 60. The second term of a regular Taylor series is equal to
zero after taking expectations as it is multiplied with E [V − E [V ]].
5 Portfolio-Selection including Hedge Funds 179
5.2 are equal to zero.7 The expected utility becomes
E [U (V )] = U (E [V ]) + 12
∂2U(V )∂V 2 E [V ] E [V − E [V ]]2 . (5.3)
So when investors decide in accordance to such a quadratic utility function, they
do not account for the third and higher central moments of the return distribution
when maximizing expected utility. To consider the mean and variance of returns is
therefore an adequate approach under the assumption of quadratic utility functions.8
�������������� �����������������
����������������
������
�����������
������������
�����
�����������
������������
�����
���� �����
������ ��
�����������
������������
���� ��
������������
����������������������������
Figure 5.1: Assumptions that lead to traditional portfolio selection
Another justification that leads to traditional portfolio selection can be found in
the distribution of returns, either on the level of the individual securities or on the
portfolio level. If the distribution of portfolio returns is completely determined by
its mean and variance, then the higher order terms in Equation 5.2 can be expressed
as functions of the first and second moment.9 In this case the mean-variance model
7 See for example Jean (1971), p. 506.8 For the quality of the approximation with Taylor’s series see for example Brockett/Garven
(1998), Hlawitschka (1994), Levy/Markowitz (1979) and Loistl (1976).
5 Portfolio-Selection including Hedge Funds 180
is valid for arbitrary preference structures.10 Examples for portfolio distributions,
which are completely described by their first two moments are the normal or the
lognormal distribution. But as we are usually interested in a portfolio of securi-
ties, this approach based on the distribution of portfolio returns oversimplifies the
problem.
When we consider the return distributions of individual securities we have to bear
an important requirement in mind. Any linear combination of the random securities
returns under consideration has to have a distribution in the same family of distribu-
tions.11 In the case of the lognormal distribution we can see that this distribution is
not closed under the formation of linear combinations.12 Therefore, the traditional
portfolio selection process starts with the means and variances of different available
securities is not valid with the assumption of lognormally distributed asset returns
as the portfolio distribution will usually not be lognormal. If we apply the assump-
tion of multivariate normally distributed returns of all risky assets, we again end up
with normally distributed portfolio returns that justify the mean-variance portfolio
selection according to Markowitz.13,14
If one of these rather strong assumptions about the assets’ return distributions or
the investor’s utility function holds true, the traditional mean-variance approach to
portfolio optimization is appropriate.15
9 The problem that the simple return over one period Rt, which is subject to the restriction
−1 ≤ Rt < ∞, is modeled with continuous distributions that allow returns below 100% is
usually neglected in this context, see Section 4.4.1 for a brief discussion of this problem.10 See for example Scott/Horvath (1980), p. 915, or Tobin (1958), p. 74.11 See Ingersoll (1987), p. 104.12 See Bamberg/Dorfleitner (2002), p. 874.13 See Huang/Litzenberger (1988), pp. 61-62.14 See Appendix B in Ingersoll (1987) for a proof that mean-variance analysis is not only valid
for multivariate normal distributions of securities returns but also for the general class of mul-
tivariate elliptical distributions, see Ingersoll (1987), pp. 104-106.15 See for example the early discussion on this topic in Borch (1969), Feldstein (1969) and the
comment in Tobin (1969).
5 Portfolio-Selection including Hedge Funds 181
5.2 Problems with Traditional Portfolio-Selection
The following sections take a closer look at the assumptions underlying the classical
theory of portfolio selection from an alternative investments point of view. If the
assumptions concerning the investors utility function and the return distribution
seem too restrictive and unrealistic, then the traditional way of selecting optimal
portfolios does not really lead to a maximization of an investor’s expected utility.
In an early publication on conditions under which the traditional mean-variance
approach is valid Paul Samuelson already concludes that “in practice, where crude
approximations may be better than none, the 2-moment models may be found to
have pragmatic usefulness”.16 In order to obtain a more realistic model, we will try
to improve the approach that leads to this “crude approximation”.
At first preferences of individual investors are examined in more detail as the classical
assumption of quadratic utility functions seems very restrictive. If this assumption
is not able to justify the mean-variance approach, the return distributions of the
individual asset returns in the portfolio have to be considered. A related and crucial
point is the modeling of dependencies between these random asset returns in order to
determine multivariate returns and the resulting diversification effect in a portfolio
context.
5.2.1 Investor Preferences
When the foundations of modern portfolio selection were laid by Markowitz (1952),
the preferences for higher moments than the variance were already addressed.17
As outlined in Section 5.1, this standard approach for selecting efficient portfolios
assumes quadratic utility functions or specific return distributions.18 It therefore
16 Samuelson (1967), p. 12.17 See Markowitz (1952), pp. 90-91.
5 Portfolio-Selection including Hedge Funds 182
incorporates only the first and the second moment of return distributions and does
not take investor preferences for higher moments like skewness and kurtosis into
account.
Such quadratic utility functions are by definition unable to incorporate preference
structures for higher moments of wealth or return distributions. When modeled as
such a quadratic function of wealth or simple returns, the utility displays the unde-
sirable properties of increasing absolute (and relative) risk aversion and satiation.
Increasing absolute risk aversion does not make sense intuitively and implies that
risky assets are treated as inferior goods. The property of satiation implies that a
further increase in wealth beyond a point of satiation does decrease utility, which is
also counter-intuitive and problematic as we usually assume individuals, who prefer
more wealth to less.19
Investor preferences for higher moments and the implications for portfolio selection
were discussed by many authors.20 The risk averse investor is assumed to have
a utility function with a preference for the first moment of the return or wealth
distribution and an aversion towards the second moment, the variance. But investors
with utility functions that exhibit positive marginal utility, consistent risk aversion
and consistency of moment preference do care about higher moments. They will
have a preference for positively skewed return distributions.21 This aversion towards
18 See Meyer (1987), Levy (1987), Meyer (1989) and the overview in Wong/Au (2002) on extensions
and restrictions for mean-variance modeling and the corresponding utility maximization.19 See for example Huang/Litzenberger (1988), p. 61, Tsiang (1972), p. 355, and Copeland et al.
(2005), pp. 56-57. For a CAPM-context Rubinstein (1976) and He/Leland (1993) showed that
a representative investor must have a power utility function (where higher moments do exist) if
the returns on the market portfolio are independently and identically distributed and markets
are perfect.20 See for example the early publications Alderfer/Bierman (1970), Arditti (1967), Arditti/Levy
(1975), Francis (1975), Ingersoll (1975), Jean (1971), Jean (1973), Kane (1982),
Kraus/Litzenberger (1976), Rubinstein (1973), Simkowitz/Beedles (1978), Scott/Horvath
(1980) and Tsiang (1972) on this topic.21 See Scott/Horvath (1980), p. 917.
5 Portfolio-Selection including Hedge Funds 183
negatively skewed returns can be interpreted as willingness of an investor to trade
some of his average return or wealth for a decreased chance that he will experience
a large loss.22 Positive preference towards a positive third moment of a return
distribution,23 consistent risk aversion and strict consistency of moment preference
implies an aversion towards the fourth moment of a return distribution.24 This
can be interpreted similar to the aversion towards the second moment, measuring
the dispersion of the return distribution. Given these (theoretical) preferences for
higher moments of return distributions, the next step will be to consider the complete
distribution of asset returns (including higher moments of the return distributions)
in the process of portfolio selection in order to obtain a more realistic optimal asset
allocation.
5.2.2 Return Distributions
As described in Section 5.1, a multivariate normal distribution of asset returns,
which is completely characterized by the means and covariances, is able to justify
the classical mean-variance approach. This is the standard approach to portfolio
modeling.25 For multivariate normal distributions the implication holds that the
marginal distributions have to be (univariate) normal. Therefore, if the marginal
distributions of asset returns are non-normal, the resulting multivariate distribution
cannot be Gaussian.26
Many researchers examined the return data of hedge funds or alternative investments
in general. The results of various studies for different hedge fund strategy data sets
are very similar. For the fund of hedge fund return series analyzed in Section 4.4
22 See Harvey et al. (2003), p. 5.23 See Harvey et al. (2003) for a brief summary of empirical evidence on skewness preference.24 See the proof in Scott/Horvath (1980), p. 918.25 Alternative justifications can include the whole class of multivariate elliptical distributions, see
Ingersoll (1987), pp. 104-106.26 See for example Schlittgen/Streitberg (1999), p. 504.
5 Portfolio-Selection including Hedge Funds 184
the hypothesis of normally distributed returns is rejected for all of the fund of
hedge fund indices. Most of the index returns show significant skewness and all of
them exhibit positive excess kurtosis. Such non-normality of (unconditional) hedge
fund return distributions is for example also identified in the empirical analysis in
Amin/Kat (2003), Brooks/Kat (2001), Geman/Kharoubi (2003), Fung/Hsieh (2001)
and Kat/Lu (2002).27 Therefore, the assumption of normally distributed returns in
order to justify the mean-variance approach is usually unrealistic.28
5.2.3 Dependence Structures
Besides the preference structure and isolated (marginal) return distributions of po-
tential portfolio assets, the dependence structure of the portfolio assets plays an
important or even the major role in modeling portfolio distributions. The depen-
dence structure links the different marginal distribution and therefore determines
the diversification effect within a portfolio.
5.2.3.1 Linear Correlation
When dependence between financial assets is measured, the (linear) coefficient of
correlation as standardized covariance is a popular but also very often misunderstood
measure.29,30 Linear correlation to quantify dependence is a cornerstone of classical
27 In most of these publications log returns are tested for normality. But as pointed out in Section
4.4.1 the differences between the resulting values for the two return definitions are rather small.
Therefore, the general test results should not be affected.28 But in some articles the normality of fund of hedge fund return data is simply assumed. It
is expected that higher moments disappear with the diversification of funds even in portfolios
with a small number of target funds. See for example Lhabitant/Learned (2002), pp. 32-33,
for this problematic assumption that is not based on empirical evidence for fund of hedge fund
returns.
5 Portfolio-Selection including Hedge Funds 185
mean-variance portfolio theory, and it plays an important role in the equilibrium
capital asset pricing model and the arbitrage pricing theory. When the distribution
of asset returns in the portfolio is a multivariate normal distribution,31 the linear
correlation (matrix) is a complete summary of the dependence structure. In general
when elliptical distributions32 are employed to model multivariate portfolio returns,
the linear correlation coefficient is a natural measure of dependence.33
But as stated at the beginning of Embrechts et al. (1999): “Correlation is a minefield
for the unwary”34. When we leave the class of multivariate elliptical distributions,
the use of the correlation coefficient might lead to fallacies about the dependence
structure of the risky assets as this (simple) measure is unable to summarize non-
linear dependencies.
Some important properties of the coefficient of correlation are presented in the fol-
lowing. First of all the coefficient of correlation is only defined when the variances of
the corresponding random variables are finite. The coefficient is calculated as stan-
dardized covariance therefore it might be problematic to use correlation with very
heavy-tailed distributions.35 Correlation is invariant under linear transformations
of random variables. When the coefficient of correlation is calculated, the random
variables are “normalized” by subtracting expected values (when covariances are
29 The so-called Pearson Product Moment correlation coefficient ρ for two random variables X
and Y is calculated as
ρ (X, Y ) =Cov [X, Y ]√
V ar [X] · V ar [Y ].
30 See for example Hilbert (1998) for details on different measures of correlation.31 This is a justification for the mean-variance approach, see Section 5.1.32 For elliptical distributions the density is constant on ellipsoids. In two dimensions the contour
plots of the density surfaces would result in ellipses. See chapters 2 and 3 in Fang et al. (1990)
on multivariate elliptical distributions.33 See Embrechts et al. (2002), pp. 187-189. Potential problems with the interpretation and
usage of the correlation coefficient for (elliptical) t-distributions are addressed in Embrechts
et al. (1999), p. 69, and Embrechts et al. (2001), p. 10.34 Embrechts et al. (1999), p. 6935 See Embrechts et al. (1999), p. 71.
5 Portfolio-Selection including Hedge Funds 186
determined) and divide by the corresponding standard deviations. But the linear
coefficient is not invariant under non-linear strictly increasing transformations of
the random variables like for example the natural logarithm.36 Independent ran-
dom variables show a correlation coefficient of zero but the converse is not true. A
correlation coefficient of zero does usually not imply independence.37 In the case of
arbitrary multivariate distributions not all values in the interval [−1, +1] are attain-
able for the coefficients of correlation.38 Elliptical distributions are usually able to
incorporate correlations for the full range from −1 to +1. But for some multivari-
ate distributions the attainable interval might be quite small.39 Here the maximum
correlation, which is attained when two random variables are perfectly positively
dependent or so-called comonotonic, differs from +1. When the dependency of two
perfectly negatively dependent random variables (for example when both can be
expressed as increasing respectively decreasing deterministic functions of a single
random variable) is measured with the correlation coefficient, the value of −1 is also
not necessarily within reach.
When we have to leave the class of multivariate normal or in general elliptical
distributions, the coefficient of correlation does no longer incorporate the whole
information about the dependence structure of the random variables. Even with the
standard assumption of normally distributed individual (marginal) asset returns
the multivariate distribution is not necessarily a multivariate normal as there are
countless multivariate distributions with marginal normal distributions.41 For four
36 See Embrechts et al. (1999), pp. 184-185.37 See for example Haerdle/Simar (2003), pp. 87-88.38 This interval with the maximum absolute value for the coefficient of correlation of 1 follows
from Cauchy’s inequality applied on the random variables minus the corresponding expected
values, see Schlittgen/Streitberg (1999), p. 4, and Abramowitz/Stegun (1972), p. 11.39 See Kat (2002a), p. 4. For an illustration of maximum and minimum attainable correlations
with different bivariate distributions see Embrechts et al. (1999), p. 70.40 See Section 5.3.2.3 for details on these dependence structures.41 See for example Section 5.3.2, as the copula technique outlined in this section can be employed
to construct numerous different multivariate distributions with identical marginal distributions.
5 Portfolio-Selection including Hedge Funds 187
Gaussian copula
−2 −1 0 1 2
−2
−1
0
1
2
−2 −1 0 1 2
−2
−1
0
1
2
t−copula
Clayton copula
−2 −1 0 1 2
−2
−1
0
1
2
−2 −1 0 1 2
−2
−1
0
1
2
Frank copula
Figure 5.2: Bivariate densities with standard normal marginals40
different dependence structures the contour plots in Figure 5.2 give an impression
on the corresponding bivariate density functions with standard normal marginal
distributions.
5.2.3.2 Empirical Evidence against Linear Correlation
As the focus of this chapter is on portfolio selection, we are interested in the realis-
tic modeling of multivariate relationships between (random) asset returns. Some of
the problems with the linear coefficient of correlation as a measure of dependency
have been outlined in 5.2.3.1. Numerous studies on hedge funds and other asset
classes find strong evidence for non-linear relationships in different bi- and multi-
variate (empirical) return distributions, which have to be considered in the modeling
5 Portfolio-Selection including Hedge Funds 188
process.
In Favre/Galeano (2001) and Favre/Galeano (2002) the authors employ non-linear
local regression analysis with a broad Swiss market index to explain returns of
different HFR strategy indices. Here the non-linearities in the relationship between
these return series are obvious. The study reveals a concave profile between some
arbitrage strategies or the HFR composite index and the Swiss market index.42
Furthermore the authors find out that with most hedge fund strategies diversification
benefits disappear for extremely negative market returns.43
In Edwards/Caglayan (2001a) the correlation of hedge fund indices with the S&P
500 index is examined in bull and bear markets over the period from 1990 to 1998.
A broad range of equally- and value-weighted strategy and composite indices is
calculated from the MAR database for monthly hedge fund returns. The authors
report significant differences in (conditional) correlation coefficients for bull markets
with a positive S&P500 return and bear markets with a negative index return.44 The
hedge fund returns in bear markets are very often negative and most style indices
exhibit higher correlation in such a market environment than in a bull market. Here
the authors identify problems with the often stated diversification benefit of hedge
funds as the correlation increases when diversification is needed most.45
A similar result is presented in Lhabitant (2002). Here the monthly returns for
the CSFB/Tremont indices are analyzed over the period from 1994 to 2001. The
data is divided into subsets of up- and down-markets for indices that represent US
equities, European equities and international bonds. For this sample, the correlation
coefficient in down markets is in most cases higher than the coefficient for up markets,
42 See Favre/Galeano (2002), pp. 11 and 13.43 See Favre/Galeano (2002), p. 16.44 Bull markets in Edwards/Caglayan (2001a) are defined as months with a S&P500 return
of 1 percent or more. In bearish months the S&P500 index loses 1 percent or more, see
Edwards/Caglayan (2001a), p. 98.45 See Edwards/Caglayan (2001a), pp. 101-102.
5 Portfolio-Selection including Hedge Funds 189
again suggesting problems when investors trust in the diversification effect.46
In Jaeger (2002) the relationship between S&P500 returns and the returns from
HFR or MAR hedge fund indices is studied. Between 1990 and 2002, most hedge
fund strategies show higher coefficients of correlation in bad months than in good
months.47 But Jaeger (2002) does also identify strategies, where it is the opposite
way around. Here a slightly positive correlation of the strategy index Futures Sys-
tematic Active with the S&P500 in good months turns into a negative correlation
in bad months indicating diversification benefits. This is because investors like pos-
itive correlation when equity (or bond) markets advance and negative correlation in
declining markets.48
The correlation coefficient of EACM indices with a balanced portfolio that consists
of the S&P500 index and a bond index for the period from 1990 to 2000 is analyzed
in Schneeweis/Spurgin (2000). The difference in the correlation for the top 40 and
the bottom 40 months (top and bottom months measured according to the return of
the combined stock and bond portfolio) is very high for several hedge fund indices.
With most strategies the correlation in top months is higher than in months with
poor performance. Like in other studies this is seen as an indication for a correlation
breakdown. But like in Jaeger (2002) the authors are also able to identify a strategy
with so-called “good correlation”, which stands for a decreasing correlation from top
to bottom months.49
In Lo (2001) a so-called “phase-locking” behavior50 is examined for different hedge
fund strategies and the S&P500 index. Lo (2001) determines an asymmetric sensi-
tivity of monthly hedge fund index data to the S&P500 index in form of different
46 See Lhabitant (2002), pp. 171-172.47 In this context “bad” and “good” months refer to the development of the S&P500.48 See Jaeger (2002), pp. 123-126.49 See Schneeweis/Spurgin (2000), pp. 3-5.50 “Phase-locking” describes situations, where physical and natural phenomena that are otherwise
uncorrelated become synchronized, see Lo (2001), p. 24.
5 Portfolio-Selection including Hedge Funds 190
beta coefficients in down-markets and in up-markets. For certain strategy indices,
seemingly unrelated returns show very high beta coefficients with the S&P500 in-
dex in case of a down-market. The author concludes that the fund returns exhibit
non-linearities that are not captured by correlation coefficients and linear factor
models.51
The study by Konberg/Lindberg (2001) analyzes the returns of hedge fund indices
from the data provider HFR with respect to the performance of the S&P500 in-
dex. The authors distinguish the 40 worst, middle, and best months of S&P500
performance, and for every data set the corresponding hedge fund index return cor-
relations are determined. Most of the indices had their lowest correlation in the top
months of the S&P500 index, and the highest correlations were calculated in the
worst periods for the S&P500 index.52
The results of these and other studies indicate non-linear dependence structures
between classical portfolio assets like stocks or bonds and hedge funds. But when
we use the correlation coefficient conditional on a return level, we have to bear some
theoretical aspects in mind. The analysis of conditional correlations is problematic as
outlined for example in Boyer et al. (1997), Longin/Solnik (2001) and Kat (2002a).
To estimate correlations conditional on different values of one (or both) variables
results in different conditional correlations as the variance of the conditioned variable
changes.53 Therefore, the absolute values for up- and downside correlations have
to be handled with care. But the problem of an asymmetric correlation profile
is a generally accepted fact and should be considered in the process of portfolio
51 See Lo (2001), pp. 24-27.52 See Konberg/Lindberg (2001), pp. 25-27.53 In Ang/Chen (2002) the authors propose to measure up- and downside-correlations and the
corresponding betas relative to a multivariate normal distribution, see Ang/Chen (2002), pp.
460-463. Longin/Solnik (2001) also analyze correlations relative to the correlations of a mul-
tivariate normal distribution, see Longin/Solnik (2001), pp. 667-670. Conditional measures of
dependence for normally- and Student-distributed variables are derived in Malevergne/Sornette
(2002).
5 Portfolio-Selection including Hedge Funds 191
modeling.54
5.2.4 Summary
If neither the quadratic utility function nor the multivariate return distributions
are able to justify the mean-variance approach, we have to find alternative ways
to determine optimal portfolios.55 Improvements of the classical Markowitz mean-
variance approach to portfolio selection were suggested by several authors but only a
few of the procedures additionally account for a flexible (non-normal or more general
non-elliptical) dependence structure of the relevant assets.56 In the following sections
we will present a modified portfolio selection technique. This technique is flexible
enough to integrate different univariate return distributions for the securities to be
included in the portfolio and different dependence structures between these potential
portfolio constituents in order to determine optimal portfolios.
54 This asymmetric correlation profile excludes even the more flexible portfolio approach with
multivariate elliptical return distributions that was justified in Ingersoll (1987). The class of
multivariate elliptical distributions is not able to model asymmetric dependencies.55 Fung/Hsieh (1999a) state that a mean-variance ranking can be (empirically) justified for a
variety of utility functions and return distributions. Rankings with quadratic approximations
to power and exponential utility functions are found to be in (close) accordance with the actual
ranking that results from the expected utility of historical returns. Therefore, the question if
higher moments of return distributions are really important has to be addressed. But as pointed
out in the conclusions section of Fung/Hsieh (1999a) the mean-variance approach is not seen as
sufficient assess risk in (not-normally distributed) hedge fund returns, see Fung/Hsieh (1999a),
p. 57. The researched hedge fund data in Fung/Hsieh (1999a) is equal to the data set in
Fung/Hsieh (1997a) but unfortunately no details are available on descriptive statistics of the
(unconditional) return distributions.56 See Harvey et al. (2003) for an bayesian approach that accommodates for skewness, co-skewness
and heavy tails. Patton (2002) introduces a distribution model with flexible return distributions
for portfolio analysis. In Davies et al. (2004) the authors analyze a model for portfolio-selection
which incorporates skewness and kurtosis.
5 Portfolio-Selection including Hedge Funds 192
5.3 Modeling Multivariate Return Distributions
In this section an alternative approach to model multivariate portfolio returns is
presented. Like Markowitz in his seminal article we will begin with the second
stage of portfolio selection: starting with relevant beliefs and ending up with the
choice of a portfolio.57,58 The method we employ to model the multivariate return
distributions is visualized in Figure 5.3.
������������� ���������
���������������������������������
�������� �����������
�������������������������������
����������������� ����� �������������
��������������� ��������������
Figure 5.3: Portfolio modeling approach
We consider a portfolio with a number of n assets. For these assets the (expected)
marginal return distributions are the basis of the portfolio considerations. With
historical return data at hand there are different methods for the derivation of such
distributions. We will use a non-parametric approach with kernel densities that
57 See Markowitz (1952), p. 77.58 Therefore, we will not discuss the problem of how to estimate parameters or distributions in
order to get predictive information on the distribution of future returns.
5 Portfolio-Selection including Hedge Funds 193
is introduced in Section 5.3.1. Other methods include the fitting of parametric
distributions or for example the fitting of an Edgeworth Expansion with estimated
moments.
When the marginal distributions are estimated we consider the multivariate depen-
dence structure between the assets. This dependence structure is modeled in Section
5.3.2 with a mathematical tool, the copula. These copula functions associate differ-
ent random variables with each other by specifying the relation of their behavior.
The resulting multivariate return distribution is the starting point for the portfolio
selection process.
5.3.1 Marginal Distributions
In the portfolio framework we will regard the first four moments of the respective
asset distributions as these are of major influence. We will introduce these four mo-
ments of a distribution in Section 5.3.1.1. Section 5.3.1.2 provides a non-parametric
method to estimate univariate return distributions: the kernel density method.
In our modeling approach the assets’ return distributions are assumed to be con-
tinuous. As discrete (or simple) returns are considered this is of course only an ap-
proximation. Due to limited liability the support of these returns should be [−1,∞)
but the continuous density functions are usually defined on R.59 The cumulative
density functions for the return distributions of the n assets under consideration are
given by F1, . . . , Fn. The corresponding probability density functions of the return
distributions are f1 . . . , fn.60,61
59 Problems with different return notions and consequences for financial modeling are for example
addressed in Bamberg/Dorfleitner (2002) and Dorfleitner (2002). For the distributions under
consideration in the empirical analysis the error due to this approximation is negligible, see
Table 5.2 in Section 5.4.2.60 Here we implicitly assume that discrete returns are defined on the real line, see Sections 4.4.1
and 5.4.3 for this problem.
5 Portfolio-Selection including Hedge Funds 194
5.3.1.1 Moments of Univariate Distribution Functions
Based on the return distributions the relevant moments of these marginal return
distributions are given in general form for all assets i ∈ {1, . . . , n} in equations 5.4
– 5.7. The first two (central) moments, mean (µi) and variance (σ2i ), of these return
distributions are62
µi = EFi [R] =
∫R fi(R) dR (5.4)
and
σ2i = EFi
[(R− EFi [R]
)2]=
∫(R− µi)
2 fi(R) dR. (5.5)
Skewness, the third moment of a distribution, is calculated as standardized third
central moment according to63
SKi = EFi
[(R− EFi [R]
σi
)3]
=
∫ (R− µi
σi
)3
fi(R) dR. (5.6)
The kurtosis KUi of a continuous return distribution Fi is also derived by the stan-
dardization of a central moment:64
KUi = EFi
[(R− EFi [R]
σi
)4]
=
∫ (R− µi
σi
)4
fi(R) dR. (5.7)
61 The cumulative density function or distribution function of such a random return variable
R is given by F (R) and it is defined as F (R) = P(R ≤ R
). Therefore, with continuous
distributions
F (R) =
R∫−∞
f (x) dx.
Mathematically the probability density function f (R) of the continuous return distribution is
defined as derivative of it’s cumulative density function and should fulfill the following two
properties:
f (R) ≥ 0,∀R ∈ R
and+∞∫−∞
f(R) dR = 1,
see for example Hartung et al. (2002), p. 106.62 See for example Abramowitz/Stegun (1972), p. 928.63 See for example Abramowitz/Stegun (1972), p. 928.64 See for example Abramowitz/Stegun (1972), p. 928.
5 Portfolio-Selection including Hedge Funds 195
5.3.1.2 Kernel Densities
A non-parametric method of modeling the univariate distribution of a random vari-
able is the kernel density estimation. With a pre-defined kernel function the density
is determined using univariate observed data for the random variable. The prob-
ability density function of a random variable Xi with T realized historical values
xi,1, . . . , xi,T is determined according to
fi (x) =1
T · h
T∑t=1
K
(x− xi,t
h
). (5.8)
The variable h describes the bandwidth or window width of the kernel estimator.
This smoothing parameter will be set to the value derived in Silverman (1986) which
is65
h =
(4
3
) 15
σT− 15 (5.9)
with σ as sample standard deviation of the xi,1, . . . , xi,T . The kernel function K(.)
determines the shape of the distribution. This function has to satisfy the non-
negativity condition K(.) > 0 and has to integrate for one:66∫ ∞
−∞K (x) dx = 1. (5.10)
When the kernel function K(.) fulfills these conditions then the function fi (.) is a
probability density function as it is also non-negative and integrates for one.67 We
will use the Gaussian kernel function
K (u) =1√2π
e
(−u2
2
)(5.11)
because of the continuity and differentiability properties.68 As a result of this esti-
mation technique we obtain marginal distributions based on the data xi,1, . . . , xi,T .
65 See Silverman (1986), pp. 45-46. Alternative bandwidth values can for example be based on
the interquartile range of the xi,1, . . . , xi,T .66 See for example Petersmeier (2003), p. 99.67 See for example Petersmeier (2003), p. 99.68 See Scott (1992), p. 140, for different kernel functions.
5 Portfolio-Selection including Hedge Funds 196
On one hand the density functions fi (.) are estimated but we can also derive the
cumulative density functions Fi (.) by integration.
5.3.2 Copulas
As the marginal return distributions for the assets can be estimated with the kernel
densities we take a look at the dependence structure of returns. In this subsection we
want to introduce the copula technique to obtain multivariate distribution functions.
This technique is especially appropriate when there are strong views about the
univariate distributions of assets and when the dependence structure is no longer
linear. Copulas are a very powerful tool for modeling (financial) dependencies.69 The
first and central difficulty when we work with this mathematical tool is the choice
of an appropriate copula function in order to link different marginal distributions.
The resulting multivariate distribution gives a complete picture of the multivariate
dependencies in the model and should therefore be as realistic as possible.
In this section we will first take a look at the mathematical definition of a copula.
Afterwards some general properties of copulas are introduced in Section 5.3.2.2 and
in Section 5.3.2.3 some important copula functions are described. Section 5.3.2.4
considers the fitting of copula function to given multivariate data and finally Section
5.3.2.5 discusses the selection of the best fitting copula function.
5.3.2.1 Definitions
The copula concept, which was introduced by Sklar (1959), allows to separate the
univariate marginal distributions and the multivariate dependence structure for a
multivariate distribution.70 Here the name “copula” emphasizes the way a copula
69 An overview on applications of copulas in financial modeling is for example given in Bouye
et al. (2000).70 See Joe (1997), p. 12.
5 Portfolio-Selection including Hedge Funds 197
function “couples” a joint or multivariate distribution to its univariate margins.71
So a copula is a function that is able to link univariate marginal distributions to a
joint multivariate distribution.72
A copula is basically defined as multivariate distribution function of random vari-
ables with standard-uniform marginal distributions.73,74 Three properties such a
copula function with standard-uniformly distributed random variables ui, . . . , un
has to fulfill are75
C (u1, . . . , un) is increasing in each component ui ∈ [0, 1] , (5.12)
C (1, . . . , 1, ui, 1, . . . , 1) = ui and
C (u1, . . . , ui = 0, . . . , un) = 0,
∀i ∈ {1, . . . , n}with ui ∈ [0, 1] ,(5.13)
and ∀ (a1, . . . , an) , (b1, . . . , bn) ∈ [0, 1]n with ai ≤ bi
2∑i1=1
· · ·2∑
in=1
(−1)i1+···+in C (x1i1 , . . . , xnin) ≥ 0
where xj1 = aj and xj2 = bj,∀j ∈ {1, . . . , n} .
(5.14)
These three characteristics of copula functions ensure that the resulting function is
a multivariate cumulative probability function.
71 See Nelsen (1999), p. 15.72 For simplicity we will consider these marginal distributions to be continuous. If the marginal
distributions are not continuous a copula representation of the joint distribution is still possible
but it is no longer unique, see Embrechts et al. (2002), p. 5, or Nelsen (1999), p. 15.73 See for example Embrechts et al. (2002), p. 4.74 Standard-uniform distributions are uniform distributions on [0, 1] with cumulative density func-
tion
FU (x) =
0,when x < 0
x,when 0 ≤ x ≤ 1
1,when x > 1
and corresponding probability density function fU (x) = 1 on the interval [0, 1]. See for example
Bamberg/Baur (2002), pp. 106-107.75 See for example Embrechts et al. (2002), p. 180, and Nelsen (1999), pp. 7-8.
5 Portfolio-Selection including Hedge Funds 198
Cumulative probability density functions of continuous random variables are uni-
formly distributed on [0, 1]. As the marginal distribution functions F1, . . . , Fn for
the random variables x1, . . . , xn are uniformly distributed on [0, 1] the copula asso-
ciated with the marginal cumulative probability functions is a function on the unit
n-cube C : [0, 1]n → [0, 1] that satisfies76
F (x1, . . . , xn) = C (F1 (x1) , . . . , Fn (xn)) . (5.15)
This multivariate extension of the famous result of Sklar (1959) is the foundation
of many applications of copula theory. From the theorem of Sklar we see that
for continuous multivariate distribution functions, the univariate margins and the
multivariate dependence structure represented by C can be separated.77
If F is again an n-dimensional distribution function with (continuous) marginal
distribution functions F1, . . . , Fn and u1, . . . , un ∈ [0, 1] then the copula function C
is given by
C (u1, . . . , un) = F(F−1
1 (u1) , . . . , F−1n (un)
)(5.16)
with F−1i as the quantile or inverse cumulative density function of Fi.
78 This rela-
tionship also provides a method to construct copulas from multivariate distribution
functions.79
From the multivariate cumulative density function represented by the copula func-
tion C in Equation 5.15 we can easily derive a multivariate probability density
function:80
76 See Joe (1997), p. 13, and Nelsen (1999), p. 41.77 See Romano (2002a), p. 4.78 The quantile function or inverse cumulative density function F−1
i of a distribution function Fi
is given by
F−1i (α) = inf {x|Fi (x) ≥ α}
with α ∈ [0, 1]. See for example Joe (1997), p. 10.79 See Nelsen (1999), p. 19 and pp. 45-51, for the inversion method.
5 Portfolio-Selection including Hedge Funds 199
f (x1, . . . , xn) =∂nF (x1, . . . , xn)
∂x1, . . . , ∂xn
=n∏
i=1
(∂Fi (xi)
∂xi
)∂nC (F1 (x1) , . . . , Fn (xn))
∂x1, . . . , ∂xn
.
(5.17)
This multivariate density function for the n underlying marginal distributions is the
product of the marginal densities (the differentiated cumulative density functions)
and the differentiated copula function.
5.3.2.2 Important Copula Properties
In this section we will describe two important properties of copula functions. At
first upper and lower bounds for multivariate copula functions are presented and
then we will focus on the attractive invariance property of copulas under increasing
and continuous transformations of the marginals.
Frechet-Hoeffding bounds for joint distribution functions also apply for copulas.81
These bounds for any copula function C with all ui in [0, 1] are given by82
W (u1, . . . , un) ≤ C (u1, . . . , un) ≤ M (u1, . . . , un) , (5.18)
with lower bound
W (u1, . . . , un) = max (u1 + . . . + un − n + 1, 0)
and upper bound
M (u1, . . . , un) = min (u1, . . . , un) .
The invariance under increasing and continuous transformations of the marginal
distributions is a very useful copula property. Strictly increasing transformation
functions T1, . . . , Tn for the underlying continuous random variables X1, . . . , Xn with
copula function C result in the transformed variables T1(X1), . . . , Tn(Xn) having the
same copula function C.83 Thus the dependence structure of X1, . . . , Xn is captured
80 See for example Romano (2002b), p. 6 for such a density function in two dimensions.81 These bounds are also known as Frechet bounds, see for example De Matteis (2001), p. 13.82 See for example Embrechts et al. (2001), pp. 4-5, and Nelsen (1999), pp. 26-27.83 See Meneguzzo/Vecchiato (2004), p. 43, and the proof in Lindskog (2000), pp. 6-7.
5 Portfolio-Selection including Hedge Funds 200
by the copula, regardless of the scale in which the variables are measured.84
5.3.2.3 Different Copula Functions
In this section we will discuss different copula functions. At first the product cop-
ula is presented for independent random variables. If the random variables are not
independent, their dependence structure can be modeled by different families and
types of copula functions. In the following we will present several parametric copula
functions from the elliptic and Archimedian class of copulas that are very common
in financial modeling.85 For these parametric copula functions a vector of copula pa-
rameters φ determines the copula function C (u1, . . . , un|φ). This section concludes
with the introduction of the empirical copula which is a non-parametric copula that
is especially useful when we will determine the best fitting copula function for a
given data set in Section 5.3.2.5.
Product Copula
When the information of independence between the random variables is given, no
dependence structure in the form of a special copula function is needed to deter-
mine the joint multivariate probability function. This link between univariate and
multivariate probability functions which is the simplest example of a copula is usu-
ally entitled ”product copula”.86 For independent random variables x1, . . . , xn the
copula function CP equals the product of the marginal cumulative density functions
F1, . . . , Fn:87
F (x1, . . . , xn) = CP (F1 (x1) , . . . , Fn (xn)) = F1 (x1) · . . . · Fn (xn) . (5.19)
84 See De Matteis (2001), p. 17.85 For an analysis of non-parametric or empirical copula functions see Durrleman et al. (2000),
pp. 7-11, and Bouye et al. (2000), pp. 22-23.86 See Nelsen (1999), p. 9.87 See for example Embrechts et al. (2002), p. 5.
5 Portfolio-Selection including Hedge Funds 201
This result is straightforward and the corresponding multivariate probability func-
tion f is the product of the given univariate density functions:
f (x1, . . . , xn) =∂nF (x1, . . . , xn)
∂x1, . . . , ∂xn
=n∏
i=1
(∂Fi (xi)
∂xi
). (5.20)
As we pointed out the assumption of independent random variables is usually not
met in the case of financial returns. Therefore, we will present further copula func-
tions that are able to produce a realistic dependence structure for the random vari-
ables under consideration.
Elliptical Copulas
As a starting point into different families of copula functions we will examine so-
called elliptical copula functions.88 To put it simple elliptical copulas are copulas
of elliptical distributions thus sharing many of their tractable properties.89 These
copulas are derived from multivariate elliptical distributions (the contour plots of
these distributions are elliptical) with Sklar’s theorem in Equation 5.15 on page 198.
The most prominent representative of this class of copulas is the Normal copula but
we will also present the t-copula that is based on the (Student) t-distribution.
The archetype of elliptical distributions is the normal distribution. The correspond-
ing copula is the Normal or Gaussian copula. This copula is parametrical as it
involves a correlation matrix ρ with the linear correlation coefficients ρi,j (see Sec-
tion 5.2.3) for every pair i, j of random variables. The n-variate Gaussian or Normal
copula CN for the quantiles u1, . . . , un which depends on the matrix ρ of coefficients
of correlation90 has the general form:91
CN (u1, . . . , un|ρ) = Φρ
(Φ−1 (u1) , . . . , Φ−1 (un)|ρ
)(5.21)
88 See Frahm et al. (2002) for a brief introduction to elliptical copulas.89 See for example Embrechts et al. (2001), pp. 22-30.90 The matrix ρ has to be symmetric and positive definite.91 See Bouye et al. (2000), p. 14.
5 Portfolio-Selection including Hedge Funds 202
where Φρ is the standardized multivariate normal distribution function and Φ−1 is
the inverse of the standardized univariate normal distribution function. The Gaus-
sian copula in Equation 5.21 is derived from the multivariate standard normal dis-
tribution. The covariance matrix for the multivariate normal distribution is equal to
the relevant (linear) correlation matrix ρ.92 Another expression for the multivariate
Normal copula with x as column vector of the variables x1, . . . , xn is therefore
CN (u1, . . . , un |ρ) =
∫ Φ−1(u1)
−∞· · ·∫ Φ−1(un)
−∞(2π)−
n2 |ρ|−
12 e[−
12xT ρ−1x]dx. (5.23)
The density of the Normal copula with correlation matrix ρ is given by:93
cN (u1, . . . , un |ρ) = |ρ|−12 e[−
12ξT ·(ρ−1−I)·ξ] (5.24)
with the column vector ξ of the normal inverse Φ−1 (u1) , . . . , Φ−1 (un).
Figure 5.4 on the following page gives an impression on how the structure of the
Gaussian copula looks like. For different coefficients of correlation the density func-
tion given in Equation 5.24 is plotted in the bivariate case.
The n-variate t-copula is derived from the multivariate (Student) t-distribution.
Similar to the Normal copula the t-copula Ct depends on the linear correlation
matrix ρ. The general form of the t-copula with ν degrees of freedom is94
Ct (u1, . . . , un|ρ, ν) = T nρ,ν
(T−1
ν (u1) , . . . , T−1ν (un)|ρ, ν
)(5.25)
where T nρ,ν denotes the cumulative density function of the n-variate t-distribution
function with correlation matrix ρ and T−1ν is the inverse of the distribution function
of the univariate t-distribution.95
92 The density of the multivariate standard normal or Gaussian distribution φ with x as column
vector of the variables x1, . . . , xn and correlation matrix ρ is:
φρ (x1, . . . , xn) = (2π)−n2 |ρ|−
12 e[−
12xT ρ−1x], (5.22)
see for example Thode (2002), pp. 182-183. For the multivariate standard normal distribution
the mean µi of all random variables xi is zero and the corresponding variances equal 1.93 See Bouye et al. (2000), p. 17.94 See Embrechts et al. (2001), pp. 26.
5 Portfolio-Selection including Hedge Funds 203
Figure 5.4: Bivariate densities of Gaussian copulas with different correlation
coefficients
The t-copula with x as column vector of the variables x1, . . . , xn can be written96
Ct (u1, . . . , un |ρ, ν ) =
∫ t−1ν (u1)
−∞· · ·∫ t−1
ν (un)
−∞
|ρ|−12
(πν)n2
Γ(ν+n2
)
Γ(ν2)
(1 +
xT ρ−1x
ν
)−ν+n2
dx,
(5.27)
with the gamma function Γ(.).97 The corresponding density function for the t-copula
95 The density of the standardized n-variate t-distribution tnR,ν with x as column vector of the
variables x1, . . . , xn and correlation matrix ρ is given by
tnρ,ν (x1, . . . , xn) = (πν)−n2 |ρ|−
12
Γ(ν+n2 )
Γ(ν2 )
(1 +
xT ρ−1xν
)− ν+n2
, (5.26)
see for example Bouye et al. (2000), p. 17.96 See Demarta/McNeil (2004), p. 3.
5 Portfolio-Selection including Hedge Funds 204
is given by
ct (u1, . . . , un|ρ, ν) = |ρ|−12
Γ(ν+n2
)[Γ(ν2)]n(1 + 1
νξT ρ−1ξ)−
ν+n2
[Γ(ν+12
)]nΓ(ν
2)
n∏i=1
(1 +ξ2i
ν)−
ν+12
, (5.29)
with ξi = t−1ν (ui), for all i ∈ {1, . . . , n}.98
Figure 5.5: Bivariate densities of t-copulas with different degrees of freedom
In Figure 5.5 the copula density is plotted for different degrees of freedom ν with a
fixed correlation while Figure 5.6 on the following page shows four t-copula densities
97 The gamma function Γ (.) is given by:
Γ (x) =∫ ∞
0
hx−1e−hdh, (5.28)
see for example Abramowitz/Stegun (1972), p. 255.98 See Bouye et al. (2000), p. 17, and Micocci/Masala (2003), p. 7.
5 Portfolio-Selection including Hedge Funds 205
with 6 degrees of freedom and different correlations. With more degrees of freedom
the probability mass in the tails of the bivariate distribution increases. When we
take a look at Figure 5.6 it is obvious that the concentration of the probability
increases with higher correlation coefficients.
Figure 5.6: Bivariate densities of t-copulas with different correlation coeffi-
cients
While elliptical copulas have many attractive features99 there are also some draw-
backs. These copulas do not have closed form expressions and they are restricted to
have radial symmetry.100 As in many circumstances it is more appropriate to model
99 For example tail dependence can be modeled with a t-copula while the Normal copula is not
able to produce such extreme dependencies, see for example Bouye et al. (2000), p. 18, and
Embrechts et al. (2001), p. 30.100 See Embrechts et al. (2001), p. 30.
5 Portfolio-Selection including Hedge Funds 206
a stronger dependence between financial losses than between gains we will also take
other copula functions into consideration.
Archimedian Copulas
In the following we present dependence functions from the Archimedian copula fam-
ily.101 In contrast to elliptical copulas these Archimedian copulas have closed form
expressions and they are not derived with the theorem of Sklar in Equation 5.15
on page 198. For Archimedian copulas the so-called generator function ϕ plays an
important role. This is a function ϕ(u) : [0, 1] → [0,∞] which is continuous, strictly
decreasing (ϕ′(u) < 0 for all u ∈ [0, 1]), and convex (ϕ′′(u) > 0 for all u ∈ [0, 1]).102
With such a generator function ϕ the function C : [0, 1]n → [0, 1]
C (u1, . . . , un) = ϕ[−1](ϕ(u1) + . . . + ϕ(un)) (5.30)
is a copula if the pseudo-inverse ϕ[−1] of ϕ is continuous, decreasing on [0,∞] and
strictly decreasing on [0, ϕ(0)].103,104
In the following we will introduce three multivariate Archimedian copula functions
that are based on different generator functions: the Clayton copula, the Frank
copula, and the Gumbel copula.
The n-variate Clayton copula has a generator ϕ(u) = u−βC − 1 with βC > 0. The
corresponding copula function CC is given by105
101 See Nelsen (1999), pp. 89-124, for a discussion of Archimedian one-parameter and two-
parameter copula functions.102 See for example Meneguzzo/Vecchiato (2004), p. 48.103 See Embrechts et al. (2001), p. 31, and Nelsen (1999), p. 90 and p. 121.104 The pseudo-inverse of ϕ is the function ϕ[−1] : [0,∞] → [0, 1] given by
ϕ[−1](t) =
{ϕ−1(t), 0 ≤ t ≤ ϕ(0)
0, ϕ(0) ≤ t ≤ ∞(5.31)
see Embrechts et al. (2001), p. 31.
5 Portfolio-Selection including Hedge Funds 207
CC (u1, . . . , un|βC) =
[n∑
i=1
u−βCi − n + 1
]− 1βC
. (5.32)
Figure 5.7: Bivariate densities of Clayton copulas with different coefficients βC
Figure 5.7 gives an impression of the dependence structure imposed by a Clayton
copula with different β coefficients.
The n-variate Frank copula is based on the generator function ϕ(u) = ln[
exp(−βF u)−1exp(−βF )−1
]with βF > 0. The multivariate Frank copula function CF for n ≥ 3 is therefore given
by106
CF (u1, . . . , un|βF ) = − 1
βF
ln
1 +
n∏i=1
(e−βF ui − 1
)(e−βF − 1)n−1
. (5.33)
In Figure 5.8 the Frank copula density is plotted for different β-values.
105 See for example Meneguzzo/Vecchiato (2004), p. 48.106 See for example Meneguzzo/Vecchiato (2004), p. 48.
5 Portfolio-Selection including Hedge Funds 208
Figure 5.8: Bivariate Densities of Frank copulas with different coefficients βF
The generator function for the n-variate Gumbel copula is ϕ(u) = (− ln u)βG with
βG ≥ 1. The corresponding multivariate Gumbel copula CF is given by107
CG (u1, . . . , un|βG) = e−
[n∑
i=1(− ln ui)
βG
] 1βG
(5.34)
Figure 5.9 on the following page shows the bivariate Gumbel copula for different
parametrizations.
Empirical Copula Functions
After introducing different parametric copula functions we will take a look at the
non-parametric empirical copula. This method for constructing a copula function
from empirical data was introduced by Deheuvels (1979).108 The empirical copula
107 See for example Romano (2002a), p. 4.
5 Portfolio-Selection including Hedge Funds 209
Figure 5.9: Bivariate densities of Gumbel copulas with different coefficients βG
function is especially useful when it comes to the selection of a particular paramet-
ric copula. In Section 5.3.2.4 we are interested in the best fit copula for particular
return data. Here we will make use of the empirical copula as a reference for the
parametric copula functions which have been fitted to the data set. We will mea-
sure the deviation of the parametric copulas from the empirical copula by different
measures of distance.
For a sample of size T with a number of n observable (random) variables the data
set for the empirical copula is given by x1,t, . . . , xn,t with t in 1, . . . , T . The corre-
sponding rank statistic for the sample is rank1,t, . . . , rankn,t with t in 1, . . . , T .109
According to Deheuvels (1981) an empirical copula is characterized as a copula C
108 Therefore, the empirical copula is often explicitly called Deheuvels copula, see Durrleman et al.
(2000), p. 7.109 See for example Hartung et al. (2002), p. 140, on rank statistics.
5 Portfolio-Selection including Hedge Funds 210
that is defined on the lattice110
L =
{(t1T
, . . . ,tnT
); ti = 1, . . . , T ; 1 ≤ i ≤ n
}(5.35)
by
C
(t1T
, . . . ,tnT
)=
1
T
T∑t=1
n∏i=1
1[ranki,t≤ti]. (5.36)
The corresponding empirical copula density or frequency c is defined in terms of the
empirical copula function C by111
c
(t1T
, . . . ,tnT
)=
2∑j1=1
. . .2∑
jn=1
(−1)j1+...+jnC
(t1 − j1 + 1
T, . . . ,
tn − jn + 1
T
). (5.37)
The straightforward relation between the empirical copula distribution C and the
empirical copula density or frequency c is given by112
C
(t1T
, . . . ,tnT
)=
t1∑j1=1
. . .tn∑
jn=1
c
(j1
T, . . . ,
jn
T
). (5.38)
Figure 5.10 illustrates the surface of an empirical copula function from two different
perspectives.
00.2
0.40.6
0.81
00.2
0.40.6
0.810
0.2
0.4
0.6
0.8
1
MSCIHFR FoF0 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
HFR FoFMSCI
Figure 5.10: Empirical copula function113
110 See Nelsen (1999), pp. 176-177, for a bivariate representation.111 See Nelsen (1999), p. 177, for the bivariate and Durrleman et al. (2000), p. 8, or Ane/Kharoubi
(2003), p. 426, for the multivariate case.112 See Durrleman et al. (2000), p. 8.113 The data for the empirical copula function is taken from the two time series MSCI and HFR
FoF, see Section 5.4.
5 Portfolio-Selection including Hedge Funds 211
5.3.2.4 Fitting Copula Functions
In this section we will introduce a method to estimate the parameters of a given cop-
ula function. Different methods to determine copula parameters have been proposed
in the literature. The maximum likelihood estimation of the parameters of a multi-
variate distribution for a given set of (multivariate) realizations is a classical method
to fit a distribution and widely used in the copula context.114 The main problem
with the maximum likelihood method is that it becomes computationally very exten-
sive in the case of increasing dimensions of the multivariate distribution.115 Another
approach for fitting copulas is to estimate measures of dependence from the data
set. According to these estimates the copula parameter is selected. This is usually
rather simple as most dependence measures can be represented with a function of
the copula under consideration.116 A major drawback of this method is that it works
only with bivariate one-parameter copula functions.117 Non-parametric estimation
of a copula function is another approach to the fitting of a multivariate distribution
to data. No particular (parametric) copula function is specified in this case, rather a
so-called empirical copula is determined.118 We will make use of the non-parametric
empirical copula that has been introduced in Section 5.3.2.3 when the fit of different
copulas is examined in Section 5.3.2.5.119
In Joe/Xu (1996) the authors proposed the method of inference functions for mar-
gins. Because of the problems with the joint estimation of the parameters of a
multivariate distribution, the copula representation is used when the parameters
114 See for example Romano (2002b), p. 6-7.115 See Durrleman et al. (2000), p. 4.116 See for example Embrechts et al. (2001) for measures of dependence for different copula func-
tions.117 See Romano (2002b), p. 9.118 See Durrleman et al. (2000), pp. 7-9.119 The construction or fitting of non-parametric copulas will not be subject of this section. See for
example Romano (2002b), p. 10, and De Matteis (2001), pp. 36-37, for different non-parametric
approaches.
5 Portfolio-Selection including Hedge Funds 212
are split into parameters for the marginal distributions and parameters for the de-
pendence structure.120 The estimation process is then conduced in two steps. At
first the parameters of the marginal distributions are estimated and then based on
the univariate marginal distributions the parameters of the multivariate dependence
structure are determined. For the estimation of the parametrical marginals and
the dependence structure Joe/Xu (1996) proposed a maximum likelihood approach.
We will make use of this idea called method of inference functions for margins as
we split the estimation process in two steps. At first the marginal distributions are
determined and afterwards the copula is fitted with a maximum likelihood approach.
Therefore, we assume that the multivariate probability function, we would like to
identify, has the density121
f (x1, . . . , xn|φ, γ1, . . . , γn) = c (F1 (x1|γ1) , . . . , Fn (xn|γn)|φ)n∏
i=1
(fi (xi|γi)). (5.39)
c(.) is the density of the copula function under consideration with parameter vec-
tor φ. The n marginal distribution functions F1, . . . , Fn and n marginal densities
f1, . . . , fn are dependent on the parameter vectors γ1, . . . , γn. This n-variate proba-
bility density function can be estimated via maximum likelihood. The log-likelihood
function for the joint distribution with a number of T observed random vectors
x1,t, . . . , xn,t with t in 1, . . . , T is122
L (φ, γ1, . . . , γn) =T∑
t=1
ln f (x1,t, . . . , xn,t|φ, γ1, . . . , γn) . (5.40)
Using Equation 5.39 the log-likelihood function in Equation 5.40 is given by123
L (φ, γ1, . . . , γn) =T∑
t=1
ln c (F1 (x1,t|γ1) , . . . , Fn (xn,t|γn)|φ) +T∑
t=1
n∑i=1
ln fi (xi,t|γi) .
(5.41)
120 See Durrleman et al. (2000), p. 4.121 See Joe/Xu (1996), p. 3.122 See Ane/Kharoubi (2003), p. 422.123 See Geman/Kharoubi (2003), p. 65.
5 Portfolio-Selection including Hedge Funds 213
The parameters γ1, . . . , γn of the marginal distributions can also be determined by a
log-likelihood estimation. If the marginal distributions with the estimated parameter
vectors γ1, . . . , γn are given the vector of copula parameters can be estimated by
maximizing the log-likelihood function over φ in order to obtain φ:124
φ = maxφ
L (φ, γ1, . . . , γn) =T∑
t=1
ln c(F1 (x1,t|γ1), . . . , Fn (xn,t|γn)|φ). (5.42)
We will employ a slightly different approach to estimate the dependence structure
for the copula function. We will use the non-parametric kernel density approach
outlined in Section 5.3.1.2 on page 195 in order to estimate the marginal densities
fi (.) and the corresponding cumulative density functions Fi (.). When the marginal
distributions are determined, the return data is transformed with the cumulative
density function of the marginals. The second step (the maximum likelihood fitting
of the copula function) is carried out as described before. The log-likelihood function
becomes
φ = maxφ
L(φ, f1, . . . , fn
)=
T∑t=1
ln c(F1 (x1,t), . . . , Fn (xn,t)|φ
). (5.43)
This approach is mainly suitable for Archimedian copulas. In case of the Normal
and the t-copula we will use the correlation matrix to fit the dependence structures.
The Gaussian copula is fully specified with the correlation data in a matrix ρ but
the t-copula additionally has the parameter ν. Therefore, in case of the t-copula
we also make use of a maximum likelihood estimation for the degrees of freedom
ν.125 With the correlation matrix held fixed we can determine the value for ν that
maximizes the probability of the data sample.
5.3.2.5 Selecting Copula Functions
In this section certain criteria for the identification of the copula function that
delivers the best fit for a given data sample is outlined. We will make use of the
124 See Joe/Xu (1996), p. 3.125 See Demarta/McNeil (2004), p. 9, for this technique.
5 Portfolio-Selection including Hedge Funds 214
empirical copula function introduced in Section 5.3.2.3. In order to compare the
difference between the parametric copulas that have been fitted with the maximum
likelihood approach in Section 5.3.2.4 and the empirical copula we can apply different
measures of distance. The empirical copula, which is used as a benchmark, is defined
on a lattice according to formula 5.35 on page 210. Therefore, the distance between
the continuous copula functions and the empirical copula is measured with discrete
norms.126
In the following we employ two measures to determine the goodness of fit, the
Anderson-Darling and the Integrated Anderson-Darling distance statistics. The
Anderson-Darling statistic (AD) is given by127
AD = max( t1
T,..., tn
T )
∣∣∣C ( t1T, . . . , tn
T
)− C
(t1T, . . . , tn
T
)∣∣∣√C(
t1T, . . . , tn
T
) (1− C
(t1T, . . . , tn
T
)) (5.44)
with ti = 1, . . . , T for all i in 1, . . . , n.
With the Anderson-Darling statistic the maximum relative128 difference between
the empirical copula function C and the parametric copula under consideration C
is determined for all the points(
t1T, . . . , tn
T
)on the multivariate lattice L. When the
parametric copula function C attains a value of 1 the denominator of Equation 5.44
becomes equal to zero. This is the case for t1 = . . . = tn = T . Therefore, we exclude
the point(
TT, . . . , T
T
)on the lattice L from the calculation of the Anderson-Darling
statistic. As the value of the empirical copula does by definition also reach 1 at
these coordinates129 the difference is still measured correctly.
The second measure for the goodness of fit for the parametric copulas is the Inte-
grated Anderson-Darling statistic (IAD) which is given by130
126 See Ane/Kharoubi (2003), p. 426.127 See Geman/Kharoubi (2003), p. 67, for the bivariate case.128 This difference is not determined in absolute terms. The difference of the distribution functions
in Equation 5.44 is calculated relative to the square root of C(
t1T , . . . , tn
T
) (1− C
(t1T , . . . , tn
T
)).
129 See Equation 5.13 on page 197.
5 Portfolio-Selection including Hedge Funds 215
IAD =T∑
t1=1
. . .
T∑tn=1
[C(
t1T, . . . , tn
T
)− C
(t1T, . . . , tn
T
)]2C(
t1T, . . . , tn
T
) (1− C
(t1T, . . . , tn
T
)) (5.45)
The Integrated Anderson-Darling statistic reflects the sum of the relative differ-
ences between the empirical copula function C and the parametric copula under
consideration C over all the points(
t1T, . . . , tn
T
)on the lattice L. Similar to the
Anderson-Darling statistic we have to exclude the coordinate(
TT, . . . , T
T
)on the
lattice L.
130 See Ane/Kharoubi (2003), p. 426, for the bivariate case.
5 Portfolio-Selection including Hedge Funds 216
5.4 Empirical Analysis
In the first sections of this chapter we discussed the theoretical background of mod-
eling multivariate dependence structures for random variables. This section outlines
the introduced copula technique for a given set of asset returns. In this single-period
model we determine different optimal asset allocations in the case of three assets,
namely bonds, stocks, and hedge funds. We discuss some rather technical matters
that concern portfolio optimization and risk measures in Section 5.4.1 and in Section
5.4.2 we take a look at the chosen portfolio constituents. Section 5.4.3 illustrates
the modeling approach for the three specific marginal distributions of asset returns
and their dependence structure. Especially the technique of fitting copula func-
tions to data and the corresponding results are presented in detail. The portfolio
optimization that is based on the scenario approach is conducted in Section 5.4.4.
Here optimal portfolio allocations are determined with respect to a variety of risk
measures.
5.4.1 Technical Aspects
In our portfolio selection framework we will consider n investment opportunities
which deliver random monthly (simple) returns R =(R1, . . . , Rn
)T
. The uncer-
tainty including the dependence structure of the random returns is modeled with
the copula approach introduced in Section 5.3. At the beginning of the period un-
der consideration the investor has to allocate his funds to n different assets. As
we already outlined in Section 5.1 the specific allocations to these investments are
denoted w1, . . . , wn and aggregated in the weight vector w. As budget constraint
these weights should sum for one. In this framework short sales are not allowed
(wi ≥ 0 for all i in 1, . . . , n) and no riskless asset is considered.
The resulting random portfolio return is a function of the portfolio weights RP (w) =
5 Portfolio-Selection including Hedge Funds 217
wTR. As we consider a general optimization in the two dimensions risk and return
it is possible to optimize different risk measures for a given (expected) portfolio
return. The common problem can be stated for an arbitrary risk measure Π which
is a function of the portfolio returns that depend on the portfolio weights w as:131
min Π (w)
subject to wTR ≥ µ
wT1 = 1
wi ≥ 0
(5.46)
with R as vector of the expected asset returns R1, . . . , Rn and an expected portfolio
return of µ.
For different risk measures Π such an optimization can be conducted. As we set
up a probabilistic model for asset returns the optimization in this framework might
be very cumbersome and time-consuming. Therefore, we employ a technique called
scenario optimization in order to work in a deterministic world where optimization
is rather simple. The scenario optimization is based on simulation from specified
probability distributions. For the generated scenarios the uncertainty is removed.
Then we are left with a deterministic problem where we have to determine an optimal
portfolio allocation in order to maximize an objective while certain constraints are
observed.132,133
In order to obtain a number of S scenarios, we will simulate from certain multivariate
distributions that are constructed with copulas. These return series are obtained by
different simulation algorithms that are based on the specific copula function under
consideration.134 The finite set of scenarios is then represented by the return vectors
131 See for example Topaloglou et al. (2002), p. 1539.132 See for example Scherer (2002), pp. 137-159, for a brief introduction to scenario optimization.133 As an alternative we could directly use multivariate distribution functions and integrate over
the different dimensions in order to determine a portfolio distribution of returns. The resulting
portfolio return distribution would be a function of the portfolio weights for which an optimiza-
tion is desired.
5 Portfolio-Selection including Hedge Funds 218
Rs = (Rs1, . . . , R
sn) with s in 1, . . . , S. Each of the S vectors comprises a particular
realization of the n asset returns which is of equal probability.135
As stated in the optimization problem 5.46 on the preceding page, risk measures
can be expressed as a function of the portfolio weights. Together with the individual
asset returns the weighting vector determines the corresponding portfolio returns.
The risk measures we employ for the following portfolio optimization analysis in
Section 5.4.4 are the standard deviation, the mean-absolute deviation, the shortfall
probability, the minimum regret, the Value-at-Risk, and the Conditional Value-at-
Risk.
The standard deviation of returns is the basic risk measure in portfolio theory. This
statistical figure gives an impression how tightly the returns are clustered around the
mean. The standard deviation (SD) is calculated as square root of the corresponding
variance of returns:
SD(w) =√
E[(RP(w)− E [RP(w)])2
]. (5.47)
In the framework with S scenarios the common risk measure from Equation 5.47
becomes
SD(w) =
√√√√1
S
S∑s=1
(wTRs −wTR
)2. (5.48)
Another risk measure that is often used as an alternative to the classical standard
deviation of returns, is the mean-absolute deviation (MAD). The MAD is defined
134 The simulation from the Gaussian copula is similar to drawing a sample from a normal distribu-
tion, see for example Romano (2002a), pp. 6-7. Like the simulation from the Gaussian copula to
determine random numbers from a multivariate t-copula involves the Cholesky decomposition
of the linear correlation matrix and in addition random variates from the χ2 distribution, see
Romano (2002b), pp. 12-13. The algorithm for random variate generation from multivariate
Archimedian copulas uses the inverse of the generator function ϕ, see Lindskog (2000), pp.
38-43.135 In a general scenario framework we could also assign different probabilities to certain scenarios,
however this is not the case here.
5 Portfolio-Selection including Hedge Funds 219
as the mean absolute deviation of a random portfolio return RP (which is a function
of the portfolio weights w) from the corresponding expected value:136
MAD(w) = E [|RP(w)− E [RP(w)]|] . (5.49)
As risk is measured as absolute deviation from the mean, outliers have less influence
on the risk measure, compared to the classical risk measure standard deviation where
squaring gives more weight to extreme values. The MAD implies that a further unit
of underperformance or outperformance relative to the mean does entail the same
difference in utility no matter how big the losses or gain already are.137 In our
scenario framework the general risk measure MAD from Equation 5.49 becomes
MAD(w) =1
S
S∑s=1
∣∣wTRs −wTR∣∣ . (5.50)
The risk measures standard deviation and mean-absolute deviation might be prob-
lematic because of the symmetry that punishes negative and positive deviations
from the mean. Therefore, we will also consider risk measures that focus only on
the downside of the return distribution.
The shortfall probability (SfP) to some defined target return is a straight forward risk
measure.138 Here an investor is directly able to infer the probability of not reaching
a specified minimum return RTarget.139 This risk figure is of special importance in the
context of asset-liability management. In our analysis we will measure the shortfall
probability relative to a return of 0%.140
SfP(w, RTarget) = P (RP(w) ≤ RTarget) . (5.51)
Another rather simple risk measure is the minimum regret (MR). Investors that
want to minimize the maximum portfolio loss should decide in accordance to this
136 See for example Topaloglou et al. (2002), p. 1542.137 See Scherer (2002), pp. 144-145.138 This risk measure is equal to the lower partial moment of order 0.139 See Roy (1952) for the so-called “safety-first” principle.140 See for example Schubert (2005) for a portfolio analysis with shortfall probabilities.
5 Portfolio-Selection including Hedge Funds 220
risk measure or at least take this measure into consideration. Therefore, decisions
based on this risk measure are usually suitable for investors with a strong form of
risk aversion.141 The minimum regret measure MR for a set of realized asset returns
is the maximal minimum portfolio return that depends on the weight vector w.142
RminP (w) = min{RP(w)} (5.52)
For the optimization in Section 5.4.4 we are interested in the maximum of these
return minima from Equation 5.52 for all the scenarios under consideration.143
The Value-at-Risk (VaR) is widely used in the financial industry to summarize risk
in a single number. With a certain level of confidence the VaR measure makes the
statement that not more than the VaR (in percent or as absolute value in a specified
currency) will be lost within a fixed period of time.144 Formally the VaR (relative
to the distribution’s expected value) is determined according to Equation 5.53:145
VaR(w, α) = E [RP(w)]− sup{x ∈ R : P(RP(w) ≤ x) ≤ α}. (5.53)
1−α is the confidence level for the VaR. In the following analysis we will work with
a confidence level of 95% which corresponds to a value of α of 5%.146
A problem with the Value-at-Risk measure is that it does not provide detailed
information on the negative tail of a return distribution as it is focused on the α-
quantile of a distribution and does not take the shape of this tail into account.147
Moreover the VaR statistic is difficult to optimize using scenarios as the VaR (like
141 See Young (1998), p. 673 and pp. 677-678.142 See for example Scherer (2002), p. 146, on the minimum regret measure.143 See Young (1998), p. 674.144 See Hull (2003), p. 346.145 See Zagst (2002), p. 252.146 The shortfall probability in Equation 5.51 with a target return that equals -VaR (measured
relative to zero) is α, see for example Deutsch (2004), p. 379.147 In Yamai/Yoshiba (2005) this potential underestimation of portfolio risk with fat tailed distri-
butions is called “tail-risk” of the VaR statistic, see Yamai/Yoshiba (2005), p. 1000.
5 Portfolio-Selection including Hedge Funds 221
the shortfall probabilities) is a non-smooth and non-convex function with respect to
the portfolio weights and exhibits multiple local extreme values.148
As the VaR measure also lacks the desirable property of subadditivity149 more so-
phisticated risk measures have been introduced. The coherent risk statistic Con-
ditional Value-at-Risk (CVaR) reports the expected value of all losses that exceed
the corresponding VaR. Therefore, we have to specify a confidence level in order to
determine the VaR. Again we will use the common confidence level of 95%. The
CVaR equals
CVaR(w, α) = −E [RP(w)|RP(w) ≤ sup{x ∈ R : P(RP(w) ≤ x) ≤ α}] . (5.54)
In order to emphasize the link between VaR and CVaR we can restate Equation
5.54:
CVaR(w, α) = −E [RP(w)|RP(w) ≤ −(VaR(w, α)− E [RP(w)])] . (5.55)
In Equation 5.55 the VaR is measured relative to the expected value.150
Despite all the problems associated with the VaR, it is the current industry standard
for measuring risk, while the CVaR gets more and more popular as a measure for
credit risk and market risk. Therefore, we will analyze both risk measures in our
portfolio framework.151
148 See Uryasev (2000), p. 15.149 See Artzner et al. (1999) for an introduction of the coherence axioms and an analysis of the
VaR.150 If the VaR is measured relative to zero we get a intuitive formula for the CVaR:
CVaR(w, α) = −E [RP(w)|RP(w) ≤ −VaR0(w, α)] . (5.56)
See for example Zagst (2002), pp. 251-266, for details on the connection between VaR and
CVaR.151 When the distribution of returns is normal, the VaR and the CVaR are equivalent in a sense
that both risk measures provide the same optimal portfolios, see for example Uryasev (2000),
p. 15.
5 Portfolio-Selection including Hedge Funds 222
The six risk measures (standard deviation, shortfall probability, minimum regret,
mean-absolute deviation, VaR and CVaR) introduced in this section will be consid-
ered in the portfolio optimization process in Section 5.4.4.
5.4.2 Portfolio Constituents
The portfolio that is analyzed should consist of three asset classes: stocks, bonds,
and hedge funds. The MSCI All Countries World Index (MSCI) is used as proxy
for the stock universe, bonds are represented by the Lehman Brothers Govern-
ment/Credit Bond Index152 (LBGC), and the fund of hedge fund index from HFR
delivers hedge fund return data. This index is abbreviated HFR FoF.153 All three
indices are available in US Dollars and the corresponding (discrete) returns are
therefore determined in this currency.
The development of the three index return series over the 96 months is illustrated
in Figure 5.11 on the next page. A total of 96 monthly index returns is used to fit
the marginal distributions and the dependence structures. The first simple returns
are from January 1997 and the last returns have been realized in December 2004.
5.4.3 Modeling Multivariate Return Distributions
5.4.3.1 Marginal Distributions
The first step to determine the multivariate distribution of returns is the modeling
of the corresponding marginal distributions. The kernel densities for the uncon-
ditional return distributions are estimated according to the method described in
152 The Lehman Brothers Government/Credit Bond Index was formerly known as Lehman Brothers
Government/Corporate Bond Index .153 As a consequence of the data problems outlined in Chapter 4 the annual returns of the HFR
FoF index were adjusted (reduced) by 0.80% p.a.
5 Portfolio-Selection including Hedge Funds 223
0 12 24 36 48 60 72 84 96−0.2
−0.1
0
0.1
0.2
MSC
I
0 12 24 36 48 60 72 84 96−0.1
−0.05
0
0.05
0.1
LB
GC
0 12 24 36 48 60 72 84 96−0.1
−0.05
0
0.05
0.1
HFR
FoF
Figure 5.11: Return series for different potential portfolio constituents
Section 5.3.1. Each marginal distribution is based on 96 monthly returns for the
corresponding index. Figure 5.12 on the following page shows the empirical cumu-
lative density functions and the cumulative density functions based on the kernel
densities.154
In order to get an idea of the general shape of the distributions Figure 5.13 illus-
trates the kernel densities over the [-0.20;0.20] interval and Table 5.1 reports certain
descriptive statistics.
The non-parametric estimate of the stock return distribution that is based on the
MSCI data is skewed to the left with an excess kurtosis close to zero. For the stocks
we would expect the highest standard deviation of returns and this is reflected in
the distribution based on the kernel density. The kernel density estimate from the
LBGC-data represents the distribution of bond returns. This density is plotted
154 In terms of the root mean squared error the cumulative kernel densities are a reasonable ap-
proximation to the empirical distribution functions for all three data sets.
5 Portfolio-Selection including Hedge Funds 224
−0.15 −0.1 −0.05 0 0.05 0.1 0.150
0.5
1
MSCI, Stocks
−0.03 −0.02 −0.01 0 0.01 0.02 0.030
0.5
1
LBGC, Bonds
−0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.080
0.5
1
HFR FoF, Hedge Funds
Figure 5.12: Empirical distribution function (black) and corresponding cumu-
lative kernel densities (grey)
Mean Mean SD Skew- Excess
p.a. ness Kurtosis
MSCI 0.61% 7.62% 4.98% -0.459 0.151
LBGC 0.54% 6.61% 1.02% -0.484 0.882
HFRI 0.57% 7.01% 1.99% -0.147 3.026
Table 5.1: Descriptive statistics for the estimated kernel densities
in the middle of Figure 5.13. A smaller return is expected while the volatility of
returns should be significantly below the volatility for the stocks. This distribution
is also skewed to the left and shows slightly more excess kurtosis compared to the
stock return distribution. The marginal distribution of the hedge fund component
is estimated with HFR FoF data. The expected monthly return on a fund of hedge
funds investment that is modeled with this kernel density is below the expected
return on stocks and above the corresponding expectation for bonds. The monthly
hedge fund return distribution has a negative skewness that is higher than the values
5 Portfolio-Selection including Hedge Funds 225
−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.20
5
10
MSCI, Stocks
−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.20
10
20
30
40
50
LBGC, Bonds
−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.20
10
20
30
HFR FoF, Hedge Funds
Figure 5.13: Marginal distributions represented by kernel densities
for stocks and bonds. While the skewness statistics favors hedge funds, the fat tails
of the estimated distribution are much more pronounced than those of bonds and
stocks. The excess kurtosis of the fund of hedge fund distribution is significantly
higher than the statistics for the other potential portfolio constituents.155
The problem with a continuous return distribution that is defined on R instead of
starting with a minimum return of −1 or −100% has to be analyzed for the fitted
kernel densities. Table 5.2 on the next page reports the cumulative probabilities
for certain return levels. All three kernel density functions assign no probability to
returns below −100%.156
155 In Table 5.1 a problem of the kernel density method becomes obvious as the moments from the
data set are not exactly met by the kernel density estimates, see Section 4.4 on the comparable
descriptive statistics of the HFR FoF return series.156 The cumulative probabilities of the three kernel distributions have been calculated at a maxi-
mum precision level of 4.9E-324.
5 Portfolio-Selection including Hedge Funds 226
Cumulative Stocks Bonds Hedge Funds
Probabilities at (MSCI) (LBGC) (HFR FoF)
-100% 0 0 0
-90% 0 0 0
-80% 4.9621E-261 0 0
-70% 8.5618E-192 0 0
-60% 5.3743E-134 0 0
-50% 1.2273E-87 0 0
Table 5.2: Cumulative probabilities for the estimated kernel densities
5.4.3.2 Multivariate Distributions
When the marginal distributions are determined we can proceed with the depen-
dence structure. The copula method to construct multivariate return distributions
that is described in Section 5.3 makes use of the dependence information in the
return series to fit a multivariate dependence structure.
MSCI LBGC HFR FoF
(Stocks) (Bonds) (Hedge Funds)
Stocks 1.00 -0.19 0.59
Bonds -0.19 1.00 -0.05
Hedge Funds 0.59 -0.05 1.00
Table 5.3: Linear correlation matrix of historical returns
As a brief summary of the dependencies the correlation matrix which is determined
from the historical return data is given in Table 5.3.157
5 Portfolio-Selection including Hedge Funds 227
3 4 5 6 7 8 9 10 11 12 13 14 15 1623
24
25
26
Degrees of Freedom
Log
-Lik
elih
ood
Figure 5.14: Log-Likelihood for different degrees of freedom ν of the t-copula
This correlation information is used with the Gaussian copula. For these linear
correlations we can furthermore determine the best fitting t-copula by maximum
likelihood.158 The results for different degrees of freedom ν are given in Figure 5.14.
We obtain a t-copula with 6 degrees of freedom as the best fit.159
For the Archimedian copula functions the copula parameters are also estimated with
the maximum likelihood approach. The estimated parameters βC , βF and βG for
the elliptical copulas are found in Table 5.4.
Clayton Frank Gumbel
β 0.2408 1.0975 1.1381
Table 5.4: Estimates for βC , βF , and βG
For the copulas introduced in Section 5.3.2 the fit relative to the empirical copula
is determined. The information on the fit of the whole dependence structure is
157 When the return data is mapped on (0,1) with the estimated cumulative kernel densities the
result is a slightly adjusted linear correlation matrix. This difference is caused by the properties
of the linear correlation coefficient, as the linear correlation is not-invariant under non-linear
transformations, see Section 5.2.3.1 for more details.158 The dependence information for elliptical copulas can also be extracted with the rank correlation
measured by Kendall’s tau, see for example Lindskog (2000) or Lindskog et al. (2001).159 The same result is obtained with Kendall’s tau as relevant correlation measure.
5 Portfolio-Selection including Hedge Funds 228
aggregated in Table 5.5. The maximum absolute difference between the empirical
copula and the different fitted copulas is reported in the second column. In the
other columns the Anderson-Darling (AD) and the Integrated Anderson-Darling
(IAD) measure of difference are given.160 For all measures the Gumbel copula is
the best fitted Archimedian dependence structure. This result is really impressive,
as the gumbel copula is fitted with one parameter while for example the t-Copula
offers four parameters (three correlation coefficients and the degrees of freedom)
to fit. While the Gumbel copula is the best overall fit in terms of the maximum
difference and the Anderson-Darling measure, the Gaussian copula gives the best
results for the Integrated Anderson-Darling measure. Furthermore the differences
between the two other goodness-of-fit measures for the elliptical copulas are rather
small. Therefore, Section 5.4.4.1 will take a closer look at the simulation results for
the Gumbel copula as an Archimedian copula function and the Gaussian copula for
the class of elliptical copulas.
Maximum AD IAD
Difference
Gaussian 0.1294 1.3257 7216.5
t-Copula (ν = 6) 0.1290 1.3256 7359.3
Gumbel 0.1248 0.2758 7335.5
Frank 0.6565 1.3928 50252
Clayton 0.9337 4.0647 840830
Table 5.5: Goodness-of-fit for the different copulas
The Anderson-Darling difference measure is illustrated in Figure 5.15. In the space
in which stocks and hedge funds are located161 the value of the fraction in equations
160 See Section 5.3.2.5.161 The third dimension in the dataset (bonds) is held constant at a value of 0.5 to plot Figure 5.15
5 Portfolio-Selection including Hedge Funds 229
5.44 and 5.45 is plotted.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Stocks uni
Hed
ge F
unds
uni
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Figure 5.15: Illustration of the AD measure for a Gumbel copula
Brighter squares in the stocks-hedge funds space indicate larger differences as de-
termined by the fraction in the Anderson-Darling measure. While the Anderson-
Darling measure reports the maximum over all (in this case three) dimensions, the
Integrated Anderson-Darling measure reports the sum of all these deviations.
For the fitted Gumbel copula with the parameter βG equal to 1.1381 Figure 5.16
displays the dependence structure in the stocks/hedge fund space.
00.2
0.40.6
0.81
0
0.2
0.4
0.6
0.8
10
1
2
3
4
5
6
7
8
Stocks uniHedge Funds uni
Figure 5.16: Dependence structure from a Gumbel copula
5 Portfolio-Selection including Hedge Funds 230
5.4.4 Portfolio Analysis
In order to determine optimal portfolio allocations the scenario optimization tech-
nique is employed for a set of 10.000 simulated random variates from two different
copula functions. As the fit for the dataset is best with the Gaussian copula (in terms
of the Integrated Anderson-Darling measure) and the Gumbel copula (in terms of
the maximum distance from the empirical copula and the Anderson-Darling mea-
sure)162 we will emphasize the simulation results for the two multivariate modeling
approaches with these copulas in the following.
After a short illustration of the scenario simulation in Section 5.4.4.1 we compare
the minimum risk portfolios for the three asset portfolios. For the six risk measures
introduced in Section 5.4.1, namely standard deviation, shortfall probability, mini-
mum regret, mean-absolute deviation, VaR, and CVaR, the minimum risk portfolios
are determined in Section 5.4.4.2. These results shall give a first impression on the
risk diversification benefits of hedge funds.
Section 5.4.4.3 analyzes the different efficient frontiers for all the risk measures and
determines the impact of the two different modeling approaches. We compare the
efficient sets for the Gaussian and the Gumbel dependence structures with respect to
the resulting risk statistics and the resulting portfolio allocations in bonds, stocks,
and hedge funds.
The benefits of hedge funds are focused in Section 5.4.4.5. In order to quantify the
diversification benefits and the return enhancement we compare the minimum risk
portfolios and additionally the complete efficient frontier with and without hedge
funds as potential portfolio ingredients.
162 See Table 5.5 on page 228 for the different goodness-of-fit measures.
5 Portfolio-Selection including Hedge Funds 231
5.4.4.1 Simulation Approach
To demonstrate the scenario simulation from the different copula functions and the
resulting multivariate distributions, the figures 5.17-5.19 visualize 1.000 draws from
a three-dimensional Gaussian copula for different bivariate dimensions. The top
left graph in Figure 5.17 shows a scatter plot for the returns of the MSCI and the
LBGC. In the top right graph the returns are mapped on the interval (0; 1) with
the cumulative density functions for the marginal kernel density estimates. The
bottom right graph shows 1000 random variates from the Gaussian copula. The
bivariate data for stocks and bonds in the bottom left plot is determined with the
inverses of the marginal cumulative density functions. The Figures 5.18 and 5.19
give further information on the multivariate dependence structure as they focus on
the stock-hedge funds and the bonds-hedge funds space. Figure 5.18 illustrates the
MSCI and HFR FoF dependence structure and the corresponding random variates
that are representative for stocks and bonds. In Figure 5.19 the same information
for the LBGC and HFR FoF is given.
−0.2 −0.1 0 0.1 0.2−0.1
−0.05
0
0.05
0.1
MSCI
LB
GC
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
MSCI uni
LB
GC
uni
−0.2 −0.1 0 0.1 0.2−0.1
−0.05
0
0.05
0.1
Stocks
Bon
ds
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
Stocks uni
Bon
ds u
ni
Figure 5.17: Bivariate scatter plots for stock and bond returns
5 Portfolio-Selection including Hedge Funds 232
−0.2 −0.1 0 0.1 0.2−0.1
−0.05
0
0.05
0.1
MSCI
HFR
FoF
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
MSCI uni
HFR
FoF
uni
−0.2 −0.1 0 0.1 0.2−0.1
−0.05
0
0.05
0.1
Stocks
Hed
ge F
unds
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
Stocks uni
Hed
ge F
unds
uni
Figure 5.18: Bivariate scatter plots for stock and hedge fund returns
−0.1 −0.05 0 0.05 0.1−0.1
−0.05
0
0.05
0.1
LBGC
HFR
FoF
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
LBGC uni
HFR
FoF
uni
−0.1 −0.05 0 0.05 0.1−0.1
−0.05
0
0.05
0.1
Bonds
Hed
ge F
unds
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
Bonds uni
Hed
ge F
unds
uni
Figure 5.19: Bivariate scatter plots for bond and hedge fund returns
5 Portfolio-Selection including Hedge Funds 233
5.4.4.2 Minimum Risk Portfolios
For the six risk measures under consideration we can determine the corresponding
minimum risk portfolios. Table 5.6 reports these portfolio allocations in stocks,
bonds, and hedge funds for the minimal risk portfolios in terms of standard de-
viation, mean-absolute deviation, shortfall probability, minimum regret, VaR, and
CVaR.
Copula Stocks Bonds Hedge Risk Expected
Funds Measure Return
SD CN 2.2% 79.4% 18.4% 0.89% 0.543%
CG 0.0% 82.3% 17.7% 0.95% 0.541%
MAD CN 0.0% 78.6% 21.4% 0.69% 0.542%
CG 0.0% 81.1% 18.9% 0.73% 0.541%
SfP CN 0.0% 81.1% 18.9% 24.5% 0.541%
CG 0.0% 84.8% 15.2% 25.6% 0.540%
MR CN 5.3% 88.0% 6.7% -3.26% 0.541%
CG 0.1% 90.1% 9.8% -3.52% 0.538%
VaR CN 5.1% 88.8% 6.1% 1.52% 0.541%
CG 0.2% 87.6% 12.2% 1.61% 0.539%
CVaR CN 2.7% 76.4% 20.9% 1.56% 0.544%
CG 2.4% 74.6% 23.1% 1.71% 0.544%
Table 5.6: Minimum risk portfolios for different risk measures
The expected returns for the minimum risk portfolios are very similar. The return
level is about 0.54%, only the expected returns for the minimum CVaR portfolios are
slightly higher with 0.544%. As the Gumbel copula is able to model more extreme
dependencies than the Gaussian copulas the reported risk measures are in all cases
higher for this Archimedian dependence structure.
5 Portfolio-Selection including Hedge Funds 234
When we compare the results for the different risk measures we obtain related port-
folio weights in case of the minimum risk portfolios for the standard deviation, the
shortfall probability, and the mean-absolute deviation. The reported portfolios for
these risk measures comprise a maximum allocation of 2% in stocks and a 79% to
85% investment in bonds. The remaining funds for hedge funds are about 15% to
21%.
The net effect of extreme returns can be evaluated when we compare the standard
deviation and the mean-absolute deviation. As the standard deviation punishes
extreme returns as the differences to the mean are squared, the minimum mean-
absolute deviation portfolios show a slightly higher concentration in the asset with
the lowest risk, namely bonds. The hedge fund quota for the mean-absolute devia-
tion portfolio is also higher than in the minimum standard deviation case.
The analysis of the minimum risk portfolios for the minimum regret and the VaR
as risk measure yields related results as both risk measures focus on the tails of the
return distributions. In case of the Gaussian dependence structure the minimum
risk investment is a 5% allocation to stocks and a 88% to 89% investment in bonds.
The remaining 6% to 7% are invested in hedge funds. For the Gumbel dependence
structure the minimum risk portfolio consist of a 10% and 12% allocation to hedge
funds and a 90% and 88% investment in bonds.
The CVaR risk measure delivers similar minimum risk portfolios in the case of the
Gumbel and the Gaussian copula. For an investor a minimum risk allocation in
terms of the CVaR an optimal allocation in our model offers the highest expected
return of all minimum risk portfolios. The allocation has a high proportion of hedge
funds in the portfolio. 21% to 23% are invested in this asset class while bonds have a
weight of 75% to 76%. The remainder of the portfolio volume is allocated to stocks.
The remaining portfolio In comparison with the minimum regret criterium or the
VaR which assign substantially lower weights to hedge funds, the CVaR does not
5 Portfolio-Selection including Hedge Funds 235
only focus on one extreme outlier in the portfolio performance or a single quantile
value but on the complete tail of the distribution.
In Figure 5.20 the different minimum risk portfolios of the multivariate models with
the Gaussian and the Gumbel copula are plotted in the standard deviation-expected
return space.
5.39 5.4 5.41 5.42 5.43 5.44 5.45
x 10−3
8.9
8.95
9
9.05
9.1
9.15
9.2x 10−3
Expected Return
Stan
dard
Dev
iatio
n (G
auss
ian)
5.38 5.39 5.4 5.41 5.42 5.43 5.44 5.45
x 10−3
9.5
9.55
9.6
9.65
9.7
9.75
9.8x 10−3
Expected Return
Stan
dard
Dev
iatio
n (G
umbe
l)
SD
MRVaR
SfP CVaRMAD
SD MADSfP
VaR
MR
CVaR
Figure 5.20: Standard deviations of all minimum risk portfolios
The standard deviations for the risk statistics MAD and SD are close in both mul-
tivariate return models as these risk measures are highly correlated.163 In terms of
the classical mean-variance portfolio selection framework most of the minimum risk
portfolios are dominated by the minimum variance (or minimum standard deviation)
portfolio.
Only the minimum CVaR portfolio offers a higher return for the higher standard
163 Only in case of a high proportion of extreme returns the optimal portfolios for standard deviation
and the mean-absolute deviation are substantially different.
5 Portfolio-Selection including Hedge Funds 236
deviation in both modeling approaches. Therefore, a classical portfolio selection
model would exclude the other minimum risk portfolios as they are not efficient in a
standard deviation-expected return sense, although these portfolios might be more
appropriate for a specific investor that should optimize one of these alternative risk
statistics.
5.4.4.3 Efficient Portfolios
The Figures 5.21-5.30 report risk-return efficient sets for the Gaussian and Gumbel
dependence structures for all of the six risk measures. Besides the risk measures
and the corresponding expected returns the information on the asset weights is also
included in these graphics.164
In case of the VaR statistic the problems addressed in Section 5.4.1 become obvious
when the efficient sets are analyzed. We obtain a non-smooth and non-convex
function with respect to the portfolio weights. Therefore, the VaR efficient sets
will not be discussed in detail and we will prefer to take a closer look at the results
for the related CVaR.
Two general structures of the weights characteristics can be identified. Both struc-
tures at first substitute bonds with hedge funds when a higher level of expected
return is targeted. The prevalent structure which results for all the risk measures
except the shortfall probability and the minimum regret further increases the ex-
posure to hedge funds and subsequently adds stocks for higher expected returns.
Therefore, the bond allocation shrinks to zero at an expected return level of approx-
imately 0.57% to 0.585%. Then the portfolio allocation is completely determined
by stocks and hedge funds according to the expected return level of the investment
alternatives.
164 The allocations are plotted in three different colors: bonds (black), stocks (light grey), and
hedge funds (dark grey).
5 Portfolio-Selection including Hedge Funds 237
In case of the shortfall probability and the minimum regret165 the weights structure
is slightly different as the allocation to bonds decreases more slowly. Because of the
risk characteristics of stocks and hedge funds the minimization of these risk measures
delivers substantially higher allocations for bonds. The hedge fund quota is at first
increased but the portfolios on the efficient set do not exhibit higher allocations than
40%. When the portfolio optimization reaches a return level of approximately 0.6%
the expected portfolio return is again completely determined by stocks and hedge
funds.
The hedge fund proportion in the portfolios is rather different for these two weighting
structures. While an investor that considers the shortfall probability as relevant
risk measure would never allocate more than 40% to hedge funds no matter what
expected return is targeted, a decision according to other risk measures could result
in a hedge fund quota of more than 80%.
The differences in the two modeling approaches are examined in Section 5.4.4.4.
The differences in the two copula approaches are evaluated and especially the weight
differences in the efficient frontier are studied. In order to isolate the diversification
benefits of hedge funds, Section 5.4.4.5 takes a look at portfolios with and without
hedge funds.
165 This structure for the minimum regret risk statistic is obtained when the uncertainty is modeled
with the Gaussian copula.
5 Portfolio-Selection including Hedge Funds 238
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.01
0.02
0.03
0.04
0.05
Expected Return
Stan
dard
Dev
iatio
n
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.21: Standard deviation - efficient frontier for Gaussian copula
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.01
0.02
0.03
0.04
0.05
Expected Return
Stan
dard
Dev
iatio
n
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.22: Standard deviation - efficient frontier for Gumbel copula
5 Portfolio-Selection including Hedge Funds 239
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
Expected Return
Mea
n−A
bsol
ute
Dev
iatio
n
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.23: MAD - efficient frontier for Gaussian copula
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
Expected Return
Mea
n−A
bsol
ute
Dev
iatio
n
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.24: MAD - efficient frontier for Gumbel copula
5 Portfolio-Selection including Hedge Funds 240
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0.2
0.25
0.3
0.35
0.4
0.45
Expected Return
Shor
tfal
l Pro
babi
lity
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.25: Shortfall probability - efficient frontier for Gaussian copula
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0.25
0.3
0.35
0.4
0.45
Expected Return
Shor
tfal
l Pro
babi
lity
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.26: Shortfall probability - efficient frontier for Gumbel copula
5 Portfolio-Selection including Hedge Funds 241
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
−0.25
−0.2
−0.15
−0.1
−0.05
0
Expected Return
Min
imum
Reg
ret
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.27: Minimum Regret - efficient frontier for Gaussian copula
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
−0.2
−0.15
−0.1
−0.05
0
Expected Return
Min
imum
Reg
ret
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.28: Minimum Regret - efficient frontier for Gumbel copula
5 Portfolio-Selection including Hedge Funds 242
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.02
0.04
0.06
0.08
0.1
0.12
Expected Return
CV
aR
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.29: CVaR - efficient frontier for Gaussian copula
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.02
0.04
0.06
0.08
0.1
0.12
Expected Return
CV
aR
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.25
0.5
0.75
1
Expected Return
Wei
ghts
Figure 5.30: CVaR - efficient frontier for Gumbel copula
5 Portfolio-Selection including Hedge Funds 243
5.4.4.4 Impact of Modeling
As mentioned before in the case of minimum risk portfolios, the Gumbel copula
delivers higher risk figures for the minimum risk portfolios. This pattern can be
revealed from figures 5.31, 5.35, 5.37, 5.33, and 5.39 that report the differences of
the risk measures for the two copula functions. But starting with an expected return
of approximately 0.55% to 0.56% most of the risk measures report higher values for
the efficient sets determined with the Gaussian copula.
As mentioned before, the Gumbel copula, which is the best fit to our three-dimensional
data set in terms of the maximum difference from the empirical copula and the
Anderson-Darling measure, is able to model extreme dependencies in a more realis-
tic way than the Gaussian copula. The dependence structure is not focused on linear
correlations and determines rather complex dependence structures. But the copula
depends on only one parameter to incorporate the dependence structure, while the
Gaussian copula is able to model the dependence in all dimensions separately with
certain correlation parameters. When several dimensions are considered and the
Gaussian copula assigns a relatively low dependence, the Gumbel copula166 models
a stronger dependence structure. This indicates less dependence in case of regular
movements for bonds and hedge funds in a Gaussian copula model, while stocks and
hedge funds or stocks and bonds exhibit less dependence in the Gumbel model.167
The weighting differences between the two modeling approaches are visualized in
figures 5.32, 5.36, 5.38, 5.34, and 5.40.168 In general the Gaussian copula model
assigns more weight to bonds and stocks than the Gumbel copula. The differences
166 As the Gumbel copula is only fitted with one parameter this dependence structure reflects an
“average” dependence on all dimensions.167 See Table 5.3 on page 226 for the correlation coefficients of the portfolio constituents.168 In order to calculate these differences the risk figures and weights for the Gumbel copula are
subtracted from the corresponding values for the Gaussian copula. The differences in the weights
are only plotted for bonds (black), and stocks (light grey). The third asset (hedge funds) in the
portfolios will change in accordance to the sum of the changes in bonds and stocks.
5 Portfolio-Selection including Hedge Funds 244
for the allocations are rather small when the standard deviation, the mean-absolute
deviation or the CVaR is considered. But for the minimum regret sizeable differences
of up to 40% for bonds and 20% for stocks are reported. This leads to efficient hedge
fund allocations with a difference of about 60% in hedge funds for the same expected
return. Again this is a result of the two different parametrization techniques of the
copulas used in the modeling approach. As the minimum regret tries to minimize
the maximum portfolio loss this risk figure is especially sensitive to the modeling of
extreme events. While the Gaussian copula is fitted with dependence information
from the linear correlation matrix the dependence of hedge funds and stocks is rather
strong. Therefore, compared to the Gumbel copula, the optimization delivers a
higher proportion of bonds in order to minimize the extreme portfolio results.169
When the risk measure shortfall probability is analyzed the resulting weights from
the two modeling approaches do not deviate substantially. The weight differences
are zero for almost the complete efficient set. As the shortfall probabilities take
all returns below a target return of zero into account, to minimize these shortfall
probabilities does not put weight on the size of extreme portfolio returns. When the
two multivariate return distribution that are based on the Gumbel and the Gaussian
copula deliver rather similar results in the “middle” of the three dimensional return-
probability space this result is no surprise. For a lower target return the differences
between the two efficient sets increase because of the deviations in the dependence
structures.
As outlined in Section 5.4.4.3 the portfolio allocations in stocks and hedge funds
above a certain expected return is completely determined by the corresponding
return expectations. Therefore, the weighting of stocks and hedge funds in the two
different copulas frameworks is equal for higher return expectations.
169 When extreme (positive or negative) returns are considered the effect of the conditional de-
pendence structure in the Gumbel copula at least partially compensates for the problem of
averaging multivariate dependencies.
5 Portfolio-Selection including Hedge Funds 245
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.01
-0.005
0
0.005
0.01
Expected Return
Ris
k D
iffe
renc
e St
anda
rd D
evia
tion
Figure 5.31: Differences in standard deviation for the efficient frontiers (Gaus-
sian copula minus Gumbel copula results)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
0
0.05
0.1
0.15
0.2
Expected Return
Wei
ghts
Dif
fere
nce
Stan
dard
Dev
iatio
n
Figure 5.32: Differences in weights for the standard deviation - efficient fron-
tiers (Gaussian copula minus Gumbel copula results)
5 Portfolio-Selection including Hedge Funds 246
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.01
-0.005
0
0.005
0.01
Expected Return
Ris
k D
iffe
renc
e M
ean-
Abs
olut
e D
evia
tion
Figure 5.33: Differences in MAD for the efficient frontiers (Gaussian copula
minus Gumbel copula results)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.1
-0.05
0
0.05
0.1
Expected Return
Wei
ghts
Dif
fere
nce
Mea
n-A
bsol
ute
Dev
iatio
n
Figure 5.34: Differences in weights for the MAD - efficient frontiers (Gaussian
copula minus Gumbel copula results)
5 Portfolio-Selection including Hedge Funds 247
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.02
-0.015
-0.01
-0.005
0
0.005
0.01
Expected Return
Ris
k D
iffe
renc
e Sh
ortf
all P
roba
bilit
y
Figure 5.35: Differences in shortfall probability for the efficient frontiers (Gaus-
sian copula minus Gumbel copula results)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Expected Return
Wei
ghts
Dif
fere
nce
Shor
tfal
l Pro
babi
lity
Figure 5.36: Differences in weights for the shortfall probability - efficient fron-
tiers (Gaussian copula minus Gumbel copula results)
5 Portfolio-Selection including Hedge Funds 248
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.01
-0.005
0
0.005
0.01
0.015
0.02
Expected Return
Ris
k D
iffe
renc
e M
inim
um R
egre
t
Figure 5.37: Differences in minimum regret for the efficient frontiers (Gaussian
copula minus Gumbel copula results)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
0
0.1
0.2
0.3
0.4
0.5
Expected Return
Wei
ghts
Dif
fere
nce
Min
imum
Reg
ret
Figure 5.38: Differences in weights for the minimum regret - efficient frontiers
(Gaussian copula minus Gumbel copula results)
5 Portfolio-Selection including Hedge Funds 249
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.01
-0.005
0
0.005
0.01
0.015
Expected Return
Ris
k D
iffe
renc
e C
VaR
Figure 5.39: Differences in CVaR for the efficient frontiers (Gaussian copula
minus Gumbel copula results)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10-3
-0.1
-0.05
0
0.05
0.1
Expected Return
Wei
ghts
Dif
fere
nce
CV
aR
Figure 5.40: Differences in weights for the CVaR - efficient frontiers (Gaussian
copula minus Gumbel copula results)
5 Portfolio-Selection including Hedge Funds 250
5.4.4.5 Benefits of Hedge Funds
After we examined the efficient sets for the three asset portfolios we analyze the
benefits of hedge fund investments. Therefore, we take a look at the risk-return
efficient sets when hedge funds are excluded from the portfolio selection process.
The resulting portfolios are presented in figures 5.41-5.45 together with the efficient
sets from Section 5.4.4.3 where investments in hedge funds were allowed.
For the minimum risk portfolios without hedge funds Table 5.7 reports the minimum
risk figures and the portfolio allocations.
Copula Stocks Bonds Risk Expected Risk
Measure Return Difference
SD CN 7.5% 92.5% 0.94% 0.541% -5.20%
CG 0.0% 100.0% 1.02% 0.535% -6.48%
MAD CN 7.0% 93.0% 0.73% 0.541% -5.48%
CG 0.0% 100.0% 0.78% 0.535% -6.41%
SfP CN 6.5% 93.5% 25.4% 0.540% -3.39%
CG 0.5% 99.5% 26.8% 0.536% -4.48%
MR CN 10.5% 89.5% -3.42% 0.543% -4.68%
CG 0.0% 100.0% -3.83% 0.535% -8.09%
VaR CN 7.0% 93.0% 1.57% 0.541% -3.18%
CG 1.0% 99.0% 1.72% 0.536% -6.40%
CVaR CN 10.0% 90.0% 1.72% 0.543% -9.30%
CG 3.0% 97.0% 1.98% 0.538% -13.64%
Table 5.7: Minimum risk portfolios without hedge funds
The column “risk difference” in Table 5.7 summarizes the net effect in terms of the
risk statistic under consideration. The negative sign indicates that hedge funds in
5 Portfolio-Selection including Hedge Funds 251
the portfolio decrease the risk measure under consideration. This is the case for
all six risk measures and both dependence structures. The effect is more distinctive
with the Gumbel copula. As the Gaussian copula assigns a high correlation to stocks
and hedge funds this is no surprise.
The highest diversification benefit in terms of the minimum risk portfolio is re-
ported for the CVaR. The minimum risk measure is decreased by 9.3% and 13.5%,
respectively. To add hedge funds to the portfolio therefore decreases the tail risk
substantially. Even the classical risk measure standard deviation is decreased by
a relative amount of 5.2% and 6.5% when hedge funds are added to the minimum
variance portfolio.
In all but one cases the return of the minimum risk portfolio is also enhanced when
hedge funds are added. This effect is small (up to 1.2% of the return without hedge
funds) but apart from the minimum regret portfolio for the Gaussian dependence
structure this result is derived for all the risk measures.
When the complete efficient sets are analyzed the impact of hedge funds on the
reduction of risk and the enhancement of expected return is again more significant
when the Gumbel model is considered. For both dependence structures the efficient
sets in terms of standard deviation, mean-absolute deviation, and CVaR are affected
most. The minimum regret efficient frontier is not that sensitive to the hedge fund
allocation although the effect in the minimum risk portfolio is substantial. When
shortfall probabilities are considered the difference in the efficient sets is also not
very severe. The reason is located in the small hedge fund allocation along the
efficient sets.170
170 See figures 5.25 and 5.26 on page 240.
5 Portfolio-Selection including Hedge Funds 252
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.01
0.02
0.03
0.04
0.05
Expected Return
Stan
dard
Dev
iatio
n (G
auss
ian)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.01
0.02
0.03
0.04
0.05
Expected Return
Stan
dard
Dev
iatio
n (G
umbe
l)
Figure 5.41: Standard deviation - efficient frontier with (black) and without
(grey) hedge funds
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.01
0.02
0.03
0.04
Expected Return
MA
D (G
auss
ian)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.01
0.02
0.03
0.04
Expected Return
MA
D (G
umbe
l)
Figure 5.42: MAD - efficient frontier with (black) and without (grey) hedge
funds
5 Portfolio-Selection including Hedge Funds 253
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0.2
0.25
0.3
0.35
0.4
0.45
Expected Return
Shor
tfal
l Pro
babi
lity
(Gau
ssia
n)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0.25
0.3
0.35
0.4
0.45
Expected Return
Shor
tfal
l Pro
babi
lity
(Gum
bel)
Figure 5.43: Shortfall probabilities - efficient frontier with (black) and without
(grey) hedge funds
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
−0.25
−0.2
−0.15
−0.1
−0.05
0
Expected Return
Min
imum
Reg
ret (
Gau
ssia
n)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
−0.2
−0.15
−0.1
−0.05
0
Expected Return
Min
imum
Reg
ret (
Gum
bel)
Figure 5.44: Minimum regret - efficient frontier with (black) and without (grey)
hedge funds
5 Portfolio-Selection including Hedge Funds 254
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.05
0.1
0.15
0.2
Expected Return
CV
aR (G
auss
ian)
5.3 5.4 5.5 5.6 5.7 5.8 5.9 6 6.1 6.2
x 10−3
0
0.05
0.1
0.15
0.2
Expected Return
CV
aR (G
umbe
l)
Figure 5.45: CVaR - efficient frontier with (black) and without (grey) hedge
funds
5.5 Summary
In this chapter we derived an alternative portfolio selection framework. This model
is able to take the distributional characteristics of hedge funds or other non-normal
investments into consideration. Because of problems with the classical portfolio
selection approach that are outlined in Section 5.2 of this chapter, we model the
multivariate distribution of asset returns with very flexible copula functions. The
method to obtain these multivariate return distributions was described in Section
5.3. The concept of copula functions we introduced allows to separate the marginal
distributions and the dependence structure for a multivariate distribution. With
this mathematical tool at hand multivariate return distributions can be modeled
with different dependence concepts. As we are interested in fitting the multivariate
distributions to data the sections 5.3.2.4 and 5.3.2.5 focus on fitting and selecting
copula functions.
5 Portfolio-Selection including Hedge Funds 255
The last section of this chapter presented an empirical example of the modeling
technique. For a portfolio with stocks, bonds and hedge funds the multivariate
return distributions were fitted to data. Based on these return distributions we
derived optimal portfolios in terms of different risk measures.
We determined minimum risk portfolios and the complete efficient frontiers for the
different risk statistics. Some of these efficient portfolios include rather high hedge
fund allocations. For the different risk measures and the different expected portfolio
returns optimal hedge fund quotas of 10% to 80% are calculated. These portfolio
weights might be too high for a practical implementation especially when regulatory
requirements for investors are considered. But as we are interested in theoretical
optimality, these results offer an indication for portfolio construction.
We analyzed the impact of two modeling approaches with the elliptical Gaussian
copula and the Archimedian Gumbel copula. The Gumbel copula is able to model
more extreme dependencies and also non-symmetrical dependencies for two random
variables. In a multidimensional setting it became obvious that the parameter of
the Archimedian copulas averages the dependencies over the different dimensions.
Here the Gaussian copula was able to model various dependence structures in dif-
ferent dimensions while the single parameter of the Gumbel copula usually over-
or underestimates bivariate dependencies. Therefore, the next step in the develop-
ment of this portfolio selection technique should be to analyze Archimedian copulas
with more parameters that are able to model extreme dependencies and different
dependence structures for different bivariate dimensions.
In Section 5.4.4.5 we finally analyzed the benefits of hedge funds in a classical bond
and stock portfolio. Therefore, we compared risk-return efficient portfolios with
and without hedge funds. As a result hedge funds enhance the portfolio return and
diversify portfolio risk for all of the examined risk statistics. The highest diversi-
fication benefits are achieved with the CVaR as relevant risk measure. Therefore,
5 Portfolio-Selection including Hedge Funds 256
especially investors that measure risk with the CVaR statistic should think about
allocating substantial proportions of their portfolio to hedge funds.
6 Summary and Conclusion
Hedge funds are not really a new capital market phenomenon but have attracted
considerable attention over the last few years. Although there are several risks as-
sociated with hedge fund investments, even the chairman of the US Federal Reserve
Board has recently come to a positive conclusion and outlook concerning them,
having emphasized the important contributions hedge funds have made to market
efficiency and financial stability by increasing market liquidity and enhancing eco-
nomic flexibility and resilience.1
Over the years there has been growing interest from investors in more complex hedge
fund strategies and reliable hedge fund data. These investor issues and the question
how hedge fund allocations can be modeled in a portfolio context form the focus of
this work.
The second chapter discussed some important hedge fund characteristics. It is clear
that the hedge fund approach to asset management is a broader and perhaps more
intuitive investment process. Therefore, hedge funds or alternative investments in
general are defined as the whole investment universe minus a small slice, namely the
traditional investments. In general the term “hedge fund” implies the full spectrum
of investment strategies. Such a broad definition indicates that hedge funds are not
homogeneous and this is the reason for the in-depth strategy analysis in this work.
The second chapter also provides an introduction to the key characteristics of hedge
1 See Greenspan (2005).
257
6 Summary and Conclusion 258
funds. Many of these characteristics are fundamental to an understanding of the
investment opportunity offered by such funds. The conclusion is drawn that the
very flexibility of hedge funds might result in a change in the investment industry
whereby what is today referred to as “alternative” will become the norm.2
Chapter 3 provides information on hedge fund trading strategies. We identify two
categories of risk, industry-inherent risk and strategy-specific risk. Using this classi-
fication we are able to separate the risk associated with the characteristics of hedge
funds, which can be seen as a systematic risk for general hedge fund investments,
from the special risk associated with a particular hedge fund trading strategies.
Based on this understanding of risk, the specific risk of different hedge fund strate-
gies is analyzed. We describe the most important trading strategies and provide
a comprehensive and detailed analysis of these strategies and the corresponding
trades. Furthermore, Chapter 3 includes a section on funds of hedge funds in which
we introduce the most important advantages and disadvantages of these investment
vehicles. In this context we also describe the key fund selection process for funds of
hedge funds. Taken as a whole, the third chapter provides an exhaustive analysis
of hedge fund strategies. Apart from legal considerations, detailed strategy analysis
is probably the most important form of information for investors willing to allocate
money to hedge funds.
The current spectrum of hedge fund indices is described in Chapter 4 where we also
identify desirable properties for hedge fund benchmarks, namely representativeness,
accuracy, transparency, and investability. Major hedge fund index providers, propri-
etary databases, and calculated indices are analyzed in order to provide an overview
of the available hedge fund data. In addition, the different biases in hedge fund
databases and indices are described, and their extent is estimated. We also present
the results of a broad range of academic studies on different databases.
2 SeeUBS (2004), p. 112.
6 Summary and Conclusion 259
As a consequence of data biases we suggest using fund of hedge funds data when
the performance of hedge fund investments in general is of interest. A detailed
data analysis of monthly returns is provided for six major fund of hedge funds
indices. We found very similar results for all indices. When the unconditional return
distributions are considered, all the historical index return distributions are skewed
and exhibit significant excess kurtosis. Based on historical data covering an eight
year period, different statistical tests reject the hypothesis of normally distributed
returns. In a time series context we find that the return series are significantly
autocorrelated. When the autocorrelation is removed from the historical return
data, we are left with a higher standard deviation of the resulting unconditional
return distribution. All in all the analyzed return distributions are highly non-
normal.
The distributional characteristics of hedge funds, the non-linear correlation struc-
tures observed in practice, and the unrealistic quadratic utility assumption lead to
a rejection of the classical Markowitz framework that is based on the first two mo-
ments of return distributions. The alternative framework that we present in Chapter
5 is based on copula theory. The chosen approach is very flexible as it allows the
modeling of marginal distributions to be separated from the corresponding depen-
dence structure. Furthermore, this approach is able to include a variety of non-linear
dependence structures. Above all, extreme dependencies between random variables
can be modeled explicitly. We present different copula functions from the elliptical
and the Archimedian class to model the multivariate dependencies.
In an empirical example we show the usefulness of the approach in practice. For
a given set of return data for bonds, stocks, and hedge funds, we apply the intro-
duced technique to model multivariate return distributions including hedge funds.
The marginal distributions are modeled with kernel densities and the dependence
structure is assumed to be Gaussian or Gumbel. To determine optimal portfolio
allocations we employ scenario optimization and a variety of risk measures.
6 Summary and Conclusion 260
All the resulting minimum risk portfolios include reasonable hedge fund allocations.
Given these optimal allocations, an implementation in a real world portfolio might
violate regulatory restrictions in some cases. However, we are primarily interested
in theoretically optimal portfolios, and the results do in any case offer indications
for portfolio construction. The optimal hedge fund allocations in these minimum
risk portfolios range from 6% to 23%. When the impact of hedge funds on the
portfolio risk and return is determined, the net effect of hedge funds is to enhance
the portfolio return and to diversify portfolio risk, no matter which dependence
structure is used. When we consider different risk measures, we obtain the best
diversification benefits of hedge funds with the CVaR statistic. Consequently, in
particular those investors who determine risk according to this risk measure should
consider investing a substantial amount in hedge funds.
Hedge funds provide a very broad field for researchers and the research presented in
this work and related publications could be extended in several directions. Hedge
funds are a very dynamic part of the investment universe, and new strategies are
constantly emerging, whether due to special market situations, research results,
or new financial products. Furthermore, new index providers and new indices or
index families are making more and more data on hedge funds available. The index
universe introduced in Chapter 4 is therefore also subject to ongoing change.
As regards the modeling approach described in Chapter 5, we could envisage other
techniques for determining the marginal distributions or the use of different copula
functions. As the fitting results for the one-parameter Archimedian copula were
good, an Archimedian copula with more than one parameter might possibly provide
an even better fit to a multidimensional data set. Another extension of our approach
might be a multi-period model.
All in all, the hedge fund industry is very dynamic. According to research institutes,
hedge funds have reached the “end of their beginning” as the industry consolidates
6 Summary and Conclusion 261
and many hedge fund managers have come to resemble traditional asset managers as
regards their organizational structures.3 Furthermore, several derivatives on hedge
funds have been introduced. Options on hedge funds are available in the OTC-
market and securitization is already on the way as a next step for this asset class.
Given the attractive return characteristics of hedge funds, portfolios of hedge funds
can serve as efficient collateral platforms for structured products and investors will
soon be able to benefit from a tranching approach to investing with a position in a
collateralized fund obligation (CFO). This development towards customizing asset-
backed tranches may stimulate yet further inflows of capital into the hedge fund
industry.4
3 See Greenwich Associates (2004).4 See for example Cheng (2002) and Mahadevan/Schwartz (2002) on this investment alternative.
Bibliography
ABN AMRO (2004), Exchange-Listed Funds of Hedge Funds, Alternatively
11(February), pp. 32–56.
Abramowitz, M./Stegun, I. (eds) (1972), Handbook of Mathematical Functions with
Formulas, Graphs, and Mathematical Tables, 10th edn, 1972, Dover Publica-
tions, New York.
Ackermann, C./McEnally, R./Ravenscraft, D. (1999), The Performance of Hedge
Funds: Risk, Return, and Incentives, Journal of Finance 54(3), pp. 832–875.
Adrian, T. (2003), Inference and Arbitrage: The Impact of Statistical Arbitrage
on Stock Prices, February, 2003, Working Paper, Massachusetts Institute of
Technology.
Agarwal, V./Daniel, N. D./Naik, N. Y. (2004), Flows, Performance, and Managerial
Incentives in the Hedge Fund Industry, May, 2004, Working Paper, Georgia
State University and London Business School.
Ahl, P. (2001), Global Macro Fund – What Lies Ahead, AIMA Newsletter
2001(April).
Ahmad, Z. (2003), Long/Short Equity Strategies, SwissHEDGE 3(2), pp. 9–14.
Ahn, H./Danilova, A./Swindle, G. (2002), Next Generation Models for Convertible
Bonds with Credit Risk, Wilmott 2002(September), pp. 78–83.
262
Bibliography 263
Ahn, H./Wilmott, P. (2003), On Exercising American Options: The Risk of Making
More Money Than You Expected, Wilmott 2003(March), pp. 52–63.
Alaganar, V./Bhar, R. (2001), Diversification gains from American Depositary Re-
ceipts and Foreign Equities: Evidence from Australian Stocks, Journal of In-
ternational Financial Markets, Institutions and Money 11(1), pp. 97–113.
Alderfer, C. P./Bierman, H. (1970), Choices with Risk: Beyond the Mean and
Variance, Journal of Business 43(3), pp. 341–353.
Alexander, C./Dimitriu, A. (2004), The Art of Investing in Hedge Funds: Fund
Selection and Optimal Allocations, 2004, Working Paper, University of Reading.
Alexander, C./Giblin, I./Weddington, W. (2001), Cointegration and Asset Alloca-
tion: A New Active Hedge Fund Strategy, 2001, Working Paper, University of
Reading.
Allen, S. L. (2003), Financial Risk Management: A Practitioner’s Guide to Manag-
ing Market and Credit Risk, 2003, John Wiley & Sons, New York et al.
Amenc, N./Curtis, S./Martellini, L. (2003), The Alpha and Omega of Hedge Fund
Performance Measurement, February, 2003, Working Paper, EDHEC Graduate
School of Business and University of Southern California.
Amenc, N./Malaise, P./Martellini, L./Sfeir, D. (2003), Tactical Style Allocation –
A New Form of Market Neutral Strategy, Journal of Alternative Investments
6(2), pp. 8–22.
Amenc, N./Malaise, P./Martellini, L./Vaissie, M. (2004), Fund of Hedge Fund Re-
porting: A Return-Based Approach to Fund of Hedge Funds Reporting, 2004,
Discussion Paper, Edhec Risk and Asset Management Research Centre.
Amenc, N./Martellini, L. (2002), The Brave New World of Hedge Fund Indexes,
2002, Working Paper, EDHEC Graduate School of Business and University of
Southern California.
Bibliography 264
Amin, G. S./Kat, H. M. (2003), Hedge Fund Performance 1990-2000: Do the “Money
Machines” Really Add Value?, Journal of Financial and Quantitative Analysis
38(2), pp. 251–274.
Amin, G./Kat, H. M. (2002a), Welcome to the Dark Side, Januar, 2002, Working
Paper, University of Reading.
Amin, G./Kat, H. M. (2002b), Who Should Buy Hedge Funds, March, 2002, Working
Paper, University of Reading.
Ammann, M./Kind, A./Wilde, C. (2003), Are Convertible Bonds Underpriced? An
Analysis of the French Market, Journal of Banking and Finance 27(4), pp. 635–
653.
Ane, T./Kharoubi, C. (2003), Dependence Structure and Risk Measure, Journal of
Business 76(3), pp. 411–438.
Ang, A./Chen, J. (2002), Asymmetric Correlations of Equity Portfolios, Journal of
Financial Economics 63(3), pp. 443–494.
Angel, J. J./Christophe, S. E./Ferri, M. G. (2003), A Close Look at Short Selling
on NASDAQ, Financial Analysts Journal 59(6), pp. 66–74.
Apfel, R. C./Parsons, J. E./Schwert, G. W./Stewart, G. S. (2001), Short Sales,
Damages and Class Certification in 10B-5 Actions, December, 2001, Working
Paper 8618, NBER.
Arditti, F. (1967), Risk and the Required Return on Equity, Journal of Finance
22(1), pp. 19–36.
Arditti, F. D./Levy, H. (1975), Portfolio Efficiency Analysis in Three Moments: The
Multiperiod Case, Journal of Finance 30(3), pp. 797–809.
Arnott, R. D./Leinweber, D. J. (1994), Long-Short Strategies Reassessed, Financial
Analysts Journal 50(5), pp. 76–78.
Bibliography 265
Artzner, P./Delbaen, F./Eber, J./Heath, D. (1999), Coherent Measures of Risk,
Mathematical Finance 9(3), pp. 203–228.
Arulpragasam, A. R./Chanos, J. S. (2000), Short Selling: A Unique Set of Risks,
in V. Reynolds Parker (ed.), Managing Hedge Fund Risk: From the Seat of the
Practitioner – Views from Investors, Counterparties, Hedge Funds and Consul-
tants, 2000, Risk Books, London, pp. 241–248.
Asness, C. (2001), Do Hedge Funds Hedge?, in Association for Investment Manage-
ment and Research (ed.), AIMR Conference Proceedings: Hedge Fund Manage-
ment, 2001, pp. 16–24.
Asness, C./Krail, R./Liew, J. (2001), Do Hedge Funds Hedge?, Journal of Portfolio
Management 28(1), pp. 6–19.
Ayache, E./Forsyth, P./Vetzal, K. (2002), Next Generation Models for Convertible
Bonds with Credit Risk, Wilmott 2002(December), pp. 68–77.
Bai, J./Ng, S. (2005), Tests for Skewness, Kurtosis, and Normality for Time Series
Data, Journal of Business Economics & Statistics 23(1), pp. 49–60.
Bailey, J. V. (1990), Some Thoughts on Performance-Based Fees, Financial Analysts
Journal 46(4), pp. 31–40.
Bamberg, G./Baur, F. (2002), Statistik, 12th edn, 2002, Oldenbourg,
Munchen/Wien.
Bamberg, G./Coenenberg, A. G. (2004), Betriebswirtschaftliche Entscheidungstheo-
rie, 12th edn, 2004, Vahlen, Munchen.
Bamberg, G./Dorfleitner, G. (2002), Is Traditional Capital Market Theory
Consistent with Fat-Tailed Log Returns?, Zeitschrift fur Betriebswirtschaft
72(8), pp. 865–878.
Banz, R. W. (1981), The Relationship Between Return and Market Value of Com-
mon Stocks, Journal of Financial Economics 9(1), pp. 3–18.
Bibliography 266
Baquero, G./Horst, J./Verbeek, M. (2002), Survival, Look-Ahead Bias and the Per-
formance of Hedge Funds, Februar, 2002, Working Paper, University of Rotter-
dam.
Bares, P./Gibson, R./Gyger, S. (2001), Style Consistency and Survival Probability
in the Hedge Funds Industry, February, 2001, Working Paper, University of
Zurich.
Barra RogersCasey (2001), An Introduction to Hedge Funds, 2001, Barra.
Basel Committee of Banking Supervision (2003), Sound Practices for the Manage-
ment and Supervision of Operational Risk, February, 2003.
Beder, T. S./Minnich, M./Shen, H./Stanton, J. (1998), Vignettes on VaR, Journal
of Financial Engineering 7(3-4), pp. 289–309.
Bekier, M. (1998), Marketing of Hedge Funds, 1998, Lang, Bern.
Benchmark Advisory (2004), Rules for the Blue Chip Hedge Fund Index (BlueX),
March, 2004.
Berenyi, Z. (2002), Measuring Hedge Fund Risk with Multi-Moment Risk Measures,
April, 2002, Working Paper, University of Munich.
Blackfish (2001), Macro is Back, Risk & Reward 2001(October), pp. 14–17.
Blanco, C./Dowd, K./Mark, R. (2005), A Liquidity Diet, FOW 409, pp. 43–46.
Blatter, J. (2002a), Convertible Bond Arbitrage, SwissHEDGE 2(2), pp. 9–13.
Blatter, J. (2002b), Merger Arbitrage, SwissHEDGE 2(2), pp. 18–22.
Blatter, J. (2003), Equity Market Neutral Strategies, SwissHEDGE 3(3), pp. 9–14.
Bookstaber, R. (2003), Hedge Fund Existential, Financial Analysts Journal
59(5), pp. 19–23.
Borch, K. (1969), A Note on Uncertainty and Indifference Curves, Review of Eco-
nomic Studies 36(105), pp. 1–4.
Bibliography 267
Bousbib, G. (2000), The Infrastructure Challenge: Empowering the Stakeholder
Through the Successful Deployment of Technology and Data, in L. Rahl (ed.),
Risk Budgeting: A New Approach to Investing, 2000, Risk Waters Group, Lon-
don, pp. 249–282.
Bouye, E./Durrleman, V./Nikeghbali, A./Riboulet, G./Roncalli, T. (2000), Copulas
for Finance: A Reading Guide and Some Applications, 2000, Working Paper,
Credit Lyonnais.
Boyer, B. H./Gibson, M. S./Loretan, M. (1997), Pitfalls in Tests for Changes in
Correlations, 1997, Working Paper, Federal Reserve System.
Brockett, P. L./Garven, J. R. (1998), A Reexamination of the Relationship Between
Preferences and Moment Ordering by Rational Risk Averse Investors, 1998,
Geneva Papers on Risk and Insurance Theory.
Brooks, C./Kat, H. M. (2001), The Statistical Properties of Hedge Fund Index Re-
turns and their Implications for Investors, Oktober, 2001, Working Paper, Uni-
versity of Reading.
Brown, S. J./Goetzmann, W. N. (1995), Performance Persistence, Journal of Fi-
nance 50(2), pp. 679–698.
Brown, S. J./Goetzmann, W. N./Ibbotson, R. G. (1999), Offshore Hedge Funds:
Survival and Performance 1989-95, Journal of Business 72(1), pp. 91–117.
Brown, S. J./Goetzmann, W. N./Liang, B. (2004), Fees on Fees in Funds of Funds,
2004, Working Paper 02-33, Yale ICF.
Brown, S. J./Goetzmann, W. N./Park, J. (2001), Careers and Survival: Compe-
tition and Risk in the Hedge Fund and CTA Industry, Journal of Finance
56(5), pp. 1869–1886.
Brown, S. J./Goetzmann, W./Ibbotson, R. G./Ross, S. A. (1992), Survivorship Bias
in Performance Studies, Review of Financial Studies 5(4), pp. 553–580.
Bibliography 268
Brunner, B./Hafner, R. (2005), Hedge Funds als Bestandteil der Strategischen Asset
Allocation?, Zeitschrift fur das gesamte Kreditwesen 58(3), pp. 129–133.
Brush, J. S. (1997), Comparisons and Combinations of Long and Long/Short Strate-
gies, Financial Analysts Journal 53(3), pp. 81–89.
Burstein, G. (1999), Macro Trading and Investment Strategies, 1999, John Wiley &
Sons, New York et al.
Bush, M. F. (2003), What is a Closed-end Fund?, February, 2003, Wachovia Securi-
ties.
Calamos, N. P. (2003), Convertible Arbitrage: Insights and Techniques for Successful
Hedging, 2003, John Wiley & Sons, New York et al.
Callaghan, S. R./Barry, C. B. (2003), Tax-Induced Trading of Equity Securities:
Evidence from the ADR Market, Journal of Finance 58(4), pp. 1583–1611.
Campell, J. Y./Lo, A. W./MacKinlay, A. C. (1997), The Econometrics of Financial
Markets, 1997, Princeton University Press, Princeton.
Campell, J. Y./Lo, A. W./MacKinlay, A. C. (1999), A Non-Random Walk Down
Wall Street, 1999, Princeton University Press, Princeton.
Campell, R./Koedijk, K./Kofman, P. (2002), Increased Correlation in Bear Markets,
Financial Analysts Journal 58(1), pp. 87–94.
Capocci, D. (2001), An Analysis of Hedge Fund Performance 1984-2000, November,
2001, Working Paper, University of Liege.
Caxton Corporation/Kingdon Capital Management/Moore Capital Manage-
ment/Soros Fund Management/Tudor Investment Corporation (2000), Sound
Practices for Hedge Fund Managers, February, 2000.
CBOT (2002), The 5-Year Term TED Spread, 2002, Chicago Board of Trade.
Chait, N. (2000), Risk Management: A Practical Approach to Managing a Portfolio
of Hedge Funds for a Large Investment Company, in V. Reynolds Parker (ed.),
Bibliography 269
Managing Hedge Fund Risk: From the Seat of the Practitioner – Views from
Investors, Counterparties, Hedge Funds and Consultants, 2000, Risk Books,
London, pp. 3–18.
Chance, D. M. (2000), Managed Futures and their Role in Investment Portfolios,
2000, Blackwell Publishers, Massachusetts.
Charoenrook, A./Daouk, H. (2003), The World Price of Short Selling, 2003, Work-
ing Paper, Cornell University.
Cheng, C. (2002), Securitization & Hedge Funds: Collateralized Fund Obligations,
2002, Working Paper, Duke University.
Citibank (1999), Depository Receipts, 1999, Citibank.
Closed-End Fund Association (2002), Understanding the Advantages of Closed-End
Funds, 2002, Closed-End Fund Association.
Conner, A. (2003), The Asset Allocation Effects of Adjusting Alternative Assets for
Stale Pricing, 2003, Working Paper, SEI Investments.
Connolly, D. (2000), CDF Playing, FOW 350, pp. 41–44.
Copeland, T. E./Weston, J. F./Shastri, K. (2005), Financial Theory and Corporate
Policy, 4th edn, 2005, Pearson Addison Wesley, Boston et al.
Cornelli, F./Li, D. D. (2001), Risk Arbitrage in Takeovers, March, 2001, Working
Paper, Duke University.
Cottier, P. (1998), Hedge Funds and Managed Futures, 2nd edn, 1998, Paul Haupt,
Bern et al.
CRA Rogers Casey (2003), Are Hedge Funds or Hedge Fund of Funds an Asset
Class or a Strategy Among Assets Classes?, 2003, Research Insights, CRA
RogersCasey.
Cruz, M./Davies, J. (2000), Operational Risk, in V. Reynolds Parker (ed.), Man-
aging Hedge Fund Risk: From the Seat of the Practitioner – Views from In-
Bibliography 270
vestors, Counterparties, Hedge Funds and Consultants, 2000, Risk Books, Lon-
don, pp. 133–140.
CSFB/Tremont (2002), CSFB/Tremont Hedge Fund Index: Construction Rules,
March, 2002.
CSFB/Tremont (2004a), CSFB/Tremont Investable Hedge Fund Index, October,
2004.
CSFB/Tremont (2004b), CSFB/Tremont Sector Invest Indices, October, 2004.
Das, N. (2003), Development of an Analytical Framework for Hedge Fund Invest-
ment, 2003, Working Paper, Bloomsberg University.
Dauenhauer, M./Gene, N./Groff, V./Yee, W./Winer, B. (2000), Tactical Trading
Strategies for Closed-End Funds, 2000, International Diversified Equity Advi-
sors.
Davies, R. J./Kat, H. M./Lu, S. (2004), Fund of Hedge Funds Portfolio Selection:
A Multiple Objective Approach, July, 2004, Working Paper, City University
London/University of Reading.
D’Avolio, G. (2002), The Market for Borrowing Stock, Journal of Financial Eco-
nomics 66(2-3), pp. 271–306.
De Matteis, R. (2001), Fitting Copulas to Data, 2001, Master Thesis, ETH Zurich.
De Souza, C./Gokcan, S. (2004), Allocation Methodologies and Customizing Hedge
Fund Multi-Manager Multi-Strategy Products, Journal of Alternative Invest-
ments 7(1), pp. 7–22.
DeBrouwer, G. (2001), Hedge Funds in Emerging Markets, 2001, Cambridge Uni-
versity Press, Cambridge.
Deheuvels, P. (1979), La fonction de dependance empirique et ses proprietes – un
test non parametrique d’independence, Academie Royale de Belgique – Bulletin
de la Classe de Science 65(5), pp. 274–292.
Bibliography 271
Deheuvels, P. (1981), A Non Parametric Test for Independence, Publications de
l’Institute de Statistique de l’Universite de Paris 26, pp. 29–50.
Demarta, S./McNeil, A. (2004), The t Copula and Related Copulas, 2004, Working
Paper, ETH Zuerich.
Deutsch, H.-P. (2004), Derivate und Interne Modelle, 3rd edn, 2004, Schaffer-
Poeschel, Stuttgart.
Deutsche Bundesbank (1999), Hedge-Fonds und ihre Rolle auf den Finanzmarkten,
March, 1999, Deutsche Bundesbank Monatsbericht.
Dorfleitner, G. (2002), Stetige versus diskrete Renditen – Uberlegungen zur richti-
gen Verwendung beider Begriffe in Theorie und Praxis, Kredit und Kapital
2002(2), pp. 216–241.
Dow Jones Indexes (2004a), Dow Jones Hedge Fund Strategy Benchmarks, February,
2004.
Dow Jones Indexes (2004b), Dow Jones Hedge Fund Strategy Benchmarks: Bench-
mark Components, November, 2004.
Dow Jones Indexes (2004c), Dow Jones Hedge Fund Strategy Benchmarks: Index
Calculation Methodology, February, 2004.
Drippe, P./Eyrick, D. (2001), Trading Strategy Forum: Merger Arbitrage, Journal
of Alternative Investments 4(2), pp. 67–72.
Duffie, D./Singleton, K. J. (2003), Credit Risk, 2003, Princeton University Press,
Princeton.
Durrleman, V./Nikeghbali, A./Roncalli, T. (2000), Which Copula is the Right One?,
2000, Working Paper, Credit Lyonnais.
EDHEC (2004), EDHEC Alternative Indexes, March, 2004.
Edwards, F. R./Caglayan, M. O. (2001a), Hedge Fund and Commodity Fund
Bibliography 272
Investments in Bull and Bear Markets, Journal of Portfolio Management
27(4), pp. 97–108.
Edwards, F. R./Caglayan, M. O. (2001b), Hedge Fund Performance and Manager
Skill, Journal of Futures Markets 21(11), pp. 1003–1028.
Edwards, F. R./Liew, J. (1999), Hedge Funds versus Managed Futures as Asset
Classes, Journal of Derivatives 6(4), pp. 45–64.
Elton, E. J./Gruber, M. J./Agrawal, D./Mann, C. (2001), Explaining the Rate
Spread on Corporate Bonds, Journal of Finance 56(1), pp. 247–277.
Embrechts, P./Lindskog, F./McNeil, A. (2001), Modelling Dependence with Copulas
and Applications to Risk Management, 2001, Working Paper, ETH Zurich.
Embrechts, P./McNeil, A. J./Straumann, D. (1999), Correlation: Pitfalls and Al-
ternatives, RISK 12(5), pp. 69–71.
Embrechts, P./McNeil, A. J./Straumann, D. (2002), Correlation and Dependence in
Risk Management: Properties and Pitfalls, in M. Dempster (ed.), Risk Manage-
ment: Value At Risk and Beyond, 2002, Cambridge University Press, pp. 176–
223.
Ennis, R. M./Sebastian, M. D. (2003), A Critical Look at the Case for Hedge Funds,
Journal of Portfolio Management 29(4), pp. 103–112.
Estenne, L. (2000), Risk Management Issues for the Family Office, in
V. Reynolds Parker (ed.), Managing Hedge Fund Risk: From the Seat of the
Practitioner – Views from Investors, Counterparties, Hedge Funds and Consul-
tants, 2000, Risk Books, London, pp. 39–48.
Fang, K.-T./Kotz, S./Ng, K. W. (1990), Symmetric Multivariate and Related Dis-
tributions, Vol. 36 of Monographs on Statistics and Applied Probability, 1990,
Chapman and Hall, London/New York.
Favre, L./Galeano, J.-A. (2001), The Inclusion of Hedge Funds in Swiss Pension
Bibliography 273
Fund Portfolios, Financial Markets and Portfolio Management 15(4), pp. 450–
472.
Favre, L./Galeano, J.-A. (2002), An Analysis of Hedge Fund Performance Using
Loess Fit Regression, Journal of Alternative Investments 4(4), pp. 8–24.
Feldstein, M. S. (1969), Mean-Variance Analysis in the Theory of Liquidity Prefer-
ence and Portfolio Selection, Review of Economic Studies 36(105), pp. 5–12.
Financial Services Authority (2002a), Hedge Funds and the FSA, August, 2002,
Discussion Paper, Financial Services Authority (UK).
Financial Services Authority (2002b), Short Selling, October, 2002, Discussion Pa-
per, Financial Services Authority (UK).
Fothergill, M./Coke, C. (2001), Funds of Hedge Funds: An Introduction to Multi-
Manager Funds, Journal of Alternative Investments 4(2), pp. 7–16.
Frahm, G./Junker, M./Szimayer, A. (2002), Elliptical Copulas: Applicability and
Limitations, 2002, Preprint, Ceasar Research Center, Bonn.
Francis, J. C. (1975), Skewness and Investors Decisions, Journal of Financial and
Quantitative Analysis 10(1), pp. 163–172.
Fredman, A. J. (2000), What you need to know about Investing in Closed-end Funds,
2000, Closed-End Fund Association.
Fredman, A. J./DeStaebler, E. L. J. (2000), Leveraged Closed-End Equity Funds:
Evaluating the Risks and Rewards, Journal of Investing 9(1), pp. 73–82.
Fredman, A. J./Scott, G. C. (1991), Investing in Closed-End Funds: Finding Value
and Building Wealth, 1991, New York Institute of Finance, New York et al.
Friedland, D. (2001), Mortgage-Backed Securities, 2001, Magnum Funds.
FTSE (2004a), FTSE Guide to Hedge Funds, July, 2004.
FTSE (2004b), Rules for the Management of FTSE Hedge, July, 2004.
Bibliography 274
Fulenwider, P. E. (1996), Fixed-Income Arbitrage Investing, in R. A. Lake (ed.),
Evaluating and Implementing Hedge Fund Strategies – The Experience of Man-
agers and Investors, 1996, Euromoney Publications, Nestor House, pp. 115–128.
Fung, H.-G./Xu, X. E./Yau, J. (2004), Do Hedge Fund Managers Display Skill?,
Journal of Alternative Investments 7(1), pp. 22–31.
Fung, W./Hsieh, D. A. (1997a), Empirical Characteristics of Dynamic Trad-
ing Strategies: The Case of Hedge Funds, Review of Financial Studies
10(2), pp. 275–302.
Fung, W./Hsieh, D. A. (1997b), Survivorship Bias and Investment Style in the
Returns of CTAs, Journal of Portfolio Management 24(1), pp. 30–41.
Fung, W./Hsieh, D. A. (1999a), Is Mean-Variance Analysis Applicable to Hedge
Funds?, Economics Letters 62(1), pp. 53–58.
Fung, W./Hsieh, D. A. (1999b), A Primer on Hedge Funds, Journal of Empirical
Finance 6(3), pp. 309–331.
Fung, W./Hsieh, D. A. (2000), Performance Characteristics of Hedge Funds and
Commodity Funds: Natural Vs. Spurious Biases, Journal of Financial and
Quantitative Analysis 35(3), pp. 291–307.
Fung, W./Hsieh, D. A. (2001), The Risk in Hedge Fund Strategies: Theory and
Evidence from Trend Followers, Review of Financial Studies 14(2), pp. 314–
341.
Fung, W./Hsieh, D. A. (2002a), Hedge-Fund Benchmarks: Information Content and
Biases, Financial Analysts Journal 58(1), pp. 22–34.
Fung, W./Hsieh, D. A. (2002b), Risk in Fixed-Income Hedge Fund Styles, The
Journal of Fixed Income 12(2), pp. 6–27.
Gabelli, M. J. (2002), Closed-End Funds, April, 2002, Gabelli & Company, Inc.
Bibliography 275
Geltner, D. (1991), Smoothing in Appraisal-Based Returns, Journal of Real Estate
Finance and Economics 4(3), pp. 327–345.
Geltner, D. (1993), Estimating Market Values from Appraised Values Without As-
suming an Efficient Market, Journal of Real Estate Research 8(3), pp. 325–345.
Geman, H./Kharoubi, C. (2003), Hedge Funds Revisited: Distributional Character-
istics, Dependence Structure and Diversification, Journal of Risk 5(4), pp. 55–
73.
Getmansky, M. (2004), The Life Cycle of Hedge Funds: Fund Flows, Size and Per-
formance, 2004, Working Paper, MIT Sloan School of Management.
Getmansky, M./Lo, A. W./Makarov, I. (2003), An Econometric Model of Serial
Correlation and Illiquidity in Hedge Fund Returns, March, 2003, Working Paper
4288-03, MIT Sloan School of Management.
GlobeOp (2003), GlobeOp Financial Services – Overview of Mission and Market
Position, in HVB Alternatives (ed.), Activity Report 2003, 2003, pp. 68–70.
Gnanadesikan, R. (1997), Methods for Statistical Data Analysis of Multivariate Ob-
servations, 2nd edn, 1997, John Wiley & Sons, New York et al.
Goetzmann, W. N./Ingersoll, J. E./Ross, S. A. (2003), High-Water Marks and Hedge
Fund Management Contracts, Journal of Finance 58(4), pp. 1685–1717.
Goricki, W./Sohnholz, D. (2004), Investierbare Hedge-Fonds-Indizes, Absolut Report
20(06/2004), pp. 30–35.
Greene, W. H. (2000), Econometric Analysis, 4th edn, 2000, Prentice Hall, New
Jersey.
Greenspan, A. (2005), Central Bank Panel Discussion: Remarks by the Chairman
Alan Greenspan, 2005, June 6, International Monetary Conference, Beijing,
The Federal Reserve Board.
Bibliography 276
Greenwich Associates (2004), Hedge Funds: The End of the Beginning, December,
2004.
Grinblatt, M./Titman, S. (1989), Mutual Fund Performance: An Analysis of Quar-
terly Portfolio Holdings, Journal of Business 62(3), pp. 393–416.
Grinold, R./Rudd, A. (1987), Incentive Fees: Who Wins? Who Loses?, Financial
Analysts Journal 43(1), pp. 27–38.
Gruenig, C. (2002), Asia Offers Attractive Opportunities for Hedge Fund Investors,
SwissHEDGE 2(1), pp. 10–13.
Haerdle, W./Simar, L. (2003), Applied Multivaiate Statistical Analysis, 2003.
Hahne, K. D./Obermann, A. (2004), Steueranderungen eroffnen Gestal-
tungsmoglichkeiten bei Wertpapier-Leihgeschaften, Die Bank 2004(3), pp. 194–
197.
Hamilton, J. D. (1994), Time Series Analysis, 1994, Princeton University Press,
Princeton, New Jersey.
Harding, D./Nakou, G. (2004), The Art of Investing in Hedge Funds, Commodities
Now 2004(March), pp. 60–64.
Hartung, J./Elpelt, B./Klosener, K.-H. (2002), Statistik, 13th edn, 2002, Olden-
bourg, Munchen/Wien.
Harvey, C. R./Liechty, J. C./Liechty, M. W./Mueller, P. (2003), Portfolio Selection
with Higher Moments, 2003, Working Paper, Duke University/National Bureau
of Economic Research/Pennsylvania State University/University of Texas.
He, H./Leland, H. (1993), On Equilibrium Asset Price Processes, Review of Finan-
cial Studies 6(3), pp. 593–617.
Hennessee, L. E./Gradante, C. J. (2002), Direct Investing in Hedge Funds versus
Fund of Hedge Funds Products, in B. B. Bruce (ed.), Hedge Fund Strategies –
A Global Outlook, 2002, Institutional Investor, New York, pp. 118–123.
Bibliography 277
HFR (2005), HFR Q1 2005 Industry Report, 2005.
Hilbert, A. (1998), Zur Theorie der Korrelationsmaße, 1998, Josef Eul,
Lohmar/Koln.
Hlawitschka, W. (1994), The Empirical Nature of Taylor-Series Approximations to
Expected Utility, American Economic Review 84(3), pp. 713–719.
Ho, H. (2005), Expectations Dashed, RISK 18(5), pp. 58–59.
Horst, J. R. T./Nijman, T. E./Verbeek, M. (2001), Eliminating Look-Ahead Bias
in Evaluating Persistence in Mutual Fund Performance, Journal of Empirical
Finance 8, pp. 345–373.
Howell, M. J. (2001), Fund Age and Performance, Journal of Alternative Investments
4(2), pp. 57–60.
Hsieh, J. (2001), Merger Arbitrage: Profits, Holdings, and Impact on the Takeover
Process, 2001, Working Paper, Ohio State University.
Huang, C.-F./Litzenberger, R. H. (1988), Foundations for Financial Economics,
1988, North-Holland, New York/Amsterdam/London.
Hull, J. C. (2003), Options, Futures, and Other Derivatives, 5th edn, 2003, Prentice
Hall, New Jersey.
Ilmanen, A./Byrne, R./Gunasekera, H./Minikin, R. (2004), Which Risks Have Been
Best Rewarded?, Journal of Portfolio Management 30(2), pp. 53–57.
Ineichen, A. M. (2000), In Search of Alpha, Oktober, 2000, UBS Warburg.
Ineichen, A. M. (2001), Who´s Long?, Juli, 2001, UBS Warburg.
Ineichen, A. M. (2003), Absolute Returns: The Risk and Opportunities of Hedge
Fund Investing, 2003, John Wiley & Sons, New York et al.
Ineichen, A. M. (2004), Absolute Returns – Bubble or New Paradigm?, Swis-
sHEDGE 4(2), pp. 22–26.
Bibliography 278
Ingersoll, J. E. (1975), Multidimensional Security Pricing, Journal of Financial and
Quantitative Analysis 10(5), pp. 785–798.
Ingersoll, J. E. (1987), Theory of Financial Decision Making, 1987, Rowman &
Littlefield Publishers, Totowa/New Jersey.
Ionescu, G. (2002), Portfolio Construction Strategies Using Cointegration, 2002,
Dissertation Paper, Academy of Economic Studies, Bucarest.
Jacobs, B. I./Levy, K. N. (1995), More on Long-Short Strategies, Financial Analysts
Journal 51(2), pp. 88–90.
Jacobs, B. I./Levy, K. N. (1996), 20 Myths About Long-Short, Financial Analysts
Journal 52(5), pp. 81–85.
Jacobs, B. I./Levy, K. N. (1997), The Long and Short on Long-Short, Journal of
Investing 6(1), pp. 73–86.
Jacobs, B. I./Levy, K. N./Starer, D. (1998), On the Optimality of Long-Short Strate-
gies, Financial Analysts Journal 54(2), pp. 40–51.
Jaeger, L. (2001), The Benefits of Alternative Investment Strategies in the Institu-
tional Portfolio, 2001, Partners Group.
Jaeger, L. (2002), Managing Risk in Alternative Investment Strategies: Successful
Investing in Hedge Funds and Managed Futures, 2002, FT Prentice Hall, Lon-
don.
Jaeger, L. (2003), Renditequellen von Hedge Funds, 2003, Alternative Investment
Forum.
Jaeger, L. (2004), Hedge Fund Indices: A New Way to Invest in Absolute Return
Strategies?, AIMA Newsletter 2004(June).
Jaeger, L./Cittadini, P./Jacquemai, M. (2000), The saisGroup Futures Index (sGFI)
– A New Passive Futures Investment Strategy, 2000, saisGroup.
Bibliography 279
Jaeger, R. A. (2000), Fund of Funds: Risk: Defining It, Measuring It and Managing
It, in V. Reynolds Parker (ed.), Managing Hedge Fund Risk: From the Seat
of the Practitioner – Views from Investors, Counterparties, Hedge Funds and
Consultants, 2000, Risk Books, London, pp. 69–79.
Jaffer, S. (2000), An Overview of Alternative Investment Strategies, in E. Hehn (ed.),
Innovative Kapitalanlagekonzepte, 2000, Gabler, Wiesbaden, pp. 225–233.
Jameson, R. (2001), Who‘s Afraid of Liquidity Risk, ERisk 2001(December), pp. 1–
3.
Jean, W. H. (1971), The Extension of Portfolio Analysis to Three or More Param-
eters, Journal of Financial and Quantitative Analysis 6(1), pp. 505–515.
Jean, W. H. (1973), More on Multidimensional Portfolio Analysis, Journal of Fi-
nancial and Quantitative Analysis 8(June), pp. 475–490.
Joe, H. (1997), Multivariate Models and Dependence Concepts, Monographs on
Statistics and Applied Probability, 1997, Chapman & Hall, London et al.
Joe, H./Xu, J. J. (1996), The Estimation Method of Inference for Margins for Mul-
tivariate Models, 1996, University of British Columbia.
Jorion, P. (1997), Value at Risk –The New Benchmark for Controlling Market Risk,
1997, Irwin, Chicago et al.
Kaiser, D. G. (2004), Hedgefonds – Entmystifizierung einer Anlageklasse, Strukturen
– Chancen – Risiken, 2004, Gabler, Wiesbaden.
Kane, A. (1982), Skewness Preference and Portfolio Choice, Journal of Financial
and Quantitative Analysis 17(1), pp. 15–23.
Kao, D.-L. (2002), Battle for Alphas: Hedge Funds versus Long-Only Portfolios,
Financial Analysts Journal 58(2), pp. 16–36.
Kat, H. M. (2001), Hedge Fund Mania, 2001, Working Paper, University of Reading.
Bibliography 280
Kat, H. M. (2002a), The Dangers of Using Correlation to Measure Dependence,
2002, Working Paper, University of Reading.
Kat, H. M. (2002b), In Search of the Optimal Fund of Hedge Funds, 2002, Working
Paper, University of Reading.
Kat, H. M./Lu, S. (2002), An Excursion Into the Statistical Properties of Hedge
Funds, 2002, Working Paper, University of Reading.
Kawaller, I. G. (1997), The TED Spread, Derivatives Quarterly 3(3), pp. 46–59.
Keiter, E. (2000), Mortgage Strategies, in V. Reynolds Parker (ed.), Managing Hedge
Fund Risk: From the Seat of the Practitioner – Views from Investors, Counter-
parties, Hedge Funds and Consultants, 2000, Risk Books, London, pp. 215–230.
Kim, M./Szakmary, A. C./Mathur, I. (2000), Price Transmission Dynamics Between
ADRs and their Underlying Foreign Securities, Journal of Banking and Finance
24(8), pp. 1359–1382.
Kohler, A. (2003), Hedge Fund Indexing: A Square Peg in a Round Hole?, June,
2003, State Street Global Advisors.
Konberg, M./Lindberg, M. (2001), Hedge Funds: A Review of Historical Perfor-
mance, Journal of Alternative Investments 4(1), pp. 21–31.
Kraus, A./Litzenberger, R. H. (1976), Skewness Preference and the Valuation of
Risk Assets, Journal of Finance 31(4), pp. 1085–1101.
Krishnamurthy, A. (2002), The Bond/Old-Bond Spread, Journal of Financial Eco-
nomics 66(2-3), pp. 463–506.
Levy, H. (1987), Two-Moment Decision Models and Expected Utility Maximization:
Comment, American Economic Review 79(3), pp. 597–600.
Levy, H./Markowitz, H. M. (1979), Approximating Expected Utility by a Function
of Mean and Variance, American Economic Review 69(3), pp. 308–317.
Bibliography 281
Lhabitant, F.-S. (2002), Hedge-Funds – Myths and Limits, 2002, John Wiley & Sons,
New York et al.
Lhabitant, F.-S./Learned, M. (2002), Hedge Fund Diversification: How Much is
Enough?, Journal of Alternative Investments 5(3), pp. 23–49.
Liang, B. (1999), On the Performance of Hedge Funds, Financial Analysts Journal
55(4), pp. 72–85.
Liang, B. (2000), Hedge Funds: The Living and the Dead, Journal of Financial and
Quantitative Analysis 35(3), pp. 309–326.
Liang, B. (2001), Hedge Fund Performance: 1990-1999, Financial Analysts Journal
57(1), pp. 11–18.
Liang, B. (2003), On the Performance of Alternative Investements: CTAs, Hedge
Funds, and Fund-of-Funds, November, 2003, Working Paper, University of Mas-
sachusetts.
Lindskog, F. (2000), Modelling Dependence with Copulas, 2000, Master Thesis, ETH
Zurich.
Lindskog, F./McNeil, A./Schmock, U. (2001), Kendall’s Tau for Elliptical Distri-
bustions, 2001, Working Paper, ETH Zurich.
Livonius, H. (2004), Investmentrechtliche Rahmenbedingungen fur Hedgefonds in
Deutschland, Wertpapier-Mitteilungen 58(2), pp. 60–69.
Lo, A. W. (2001), Risk Management for Hedge Funds: Introduction and Overview,
Financial Analysts Journal 57(6), pp. 16–33.
Lo, A. W. (2002), The Statistics of Sharpe Ratios, Financial Analysts Journal
58(4), pp. 36–52.
Loistl, O. (1976), The Erroneous Approximation of Expected Utility by Means of
a Taylor´s Series Expansion: Analytic and Computational Results, American
Economic Review 66(5), pp. 904–910.
Bibliography 282
Longin, F./Solnik, B. (2001), Extreme Correlation of International Equity Markets,
Journal of Finance 56(2), pp. 649–676.
Lungarella, G. (2002), Managed Futures: A Real Alternative, SwissHEDGE
2(4), pp. 9–13.
Lungarella, G. (2004), Investing in Global Macro, SwissHEDGE 4(1), pp. 10–14.
Mader, W./Echter, C. (2004), Der Markt fur Hedgefonds-Zertifikate in Deutschland,
FinanzBetrieb 6(9), pp. 612–619.
Mahadevan, S./Schwartz, D. (2002), Hedge Fund Collateralized Fund Obligations,
Journal of Alternative Investments 5(3), pp. 45–62.
Maier, J./Brown, J. (2000), All about Closed-end Funds, March, 2000, PaineWebber.
Malevergne, Y./Sornette, D. (2002), Investigating Extreme Dependences: Concepts
and Tools, March, 2002, University of Nice/University of Lyon.
Malkiel, B. (1995), Returns from Investing in Mutual Funds 1971 to 1991, Journal
of Finance 50(2), pp. 549–572.
Markowitz, H. (1952), Portfolio Selection, Journal of Finance 7(1), pp. 77–91.
Mattinson, A. (2002), CDFs Make the Difference, Capital Markets
2002(Q1), pp. 83–87.
Meneguzzo, D./Vecchiato, W. (2004), Copula Sensitivity in Collateralized
Debt Obligations and Basket Default Swaps, Journal of Futures Markets
24(1), pp. 37–70.
Mercer Oliver Wyman (2005), Das Ende der Warm-Up Phase, February, 2005.
Meyer, J. (1987), Two-Moment Decision Models and Expected Utility Maximization,
American Economic Review 77(3), pp. 421–430.
Meyer, J. (1989), Two-Moment Decision Models and Expected Utility Maximization:
Reply, American Economic Review 79(3), p. 603.
Bibliography 283
Michaud, R. (1994), Reply to Arnott and Leinweber, Financial Analysts Journal
50(5), pp. 78–81.
Michaud, R. O. (1993), Are Long-Short Equity Strategies Superior?, Financial An-
alysts Journal 49(6), pp. 44–49.
Micocci, M./Masala, G. (2003), Pricing Pension Fund Guarantees Using a Copula
Approach, 2003, Working Paper, University of Cagliari.
Mishkin, F. S./Eakins, S. G. (2006), Financial Markets and Institutions, 5th edn,
2006, Pearson Addison Wesley, Boston et al.
Mitchell, M./Pulvino, T. (2001), Characteristics of Risk and Return in Risk Arbi-
trage, Journal of Finance 56(6), pp. 2135–2175.
Mogford, A. (2005), Gaining a Foothold, FOW 409, pp. 30–35.
Moix, P.-Y. (2004), Stilverschiebungen bei Hedge Fonds – Beobachtung, Erkennung
und Kontrolle – Teil 1, Absolut Report 18(02/2004), pp. 8–13.
Moore, B./O’Brien, K. (2002), How to Hire a Fund-of-Hedge-Funds Manager, Ben-
efits Canada 26(9), pp. 30–34.
MSCI (2002), MSCI Hedge Fund Index Methodology: Objectives & Guiding Princi-
ples, Classification Standard, Index Construction and Maintenance, October,
2002.
MSCI (2003a), Investable Hedge Fund Indices Methodology: Objective and Guiding
Principles, Index Construction and Maintenance, December, 2003.
MSCI (2003b), MSCI Hedge Fund Indices: Index Methodology Update, November,
2003.
Murray, M. P. (2000), Risk Management for a Distressed Securities Portfolio, in
V. Reynolds Parker (ed.), Managing Hedge Fund Risk: From the Seat of the
Practitioner – Views from Investors, Counterparties, Hedge Funds and Consul-
tants, 2000, Risk Books, London, pp. 231–240.
Bibliography 284
Nakakubo, F. (2002), Introduction to Alternative Investment Strategies – Risk Lurk-
ing Behind the Hedge Fund Boom, Oktober, 2002, NLI Research.
Nelsen, R. B. (1999), An Introduction to Copulas, Lecture Notes in Statistics 139,
1999, Springer, New York et al.
New, D. (2001), Hedge Fund Data, March, 2001, Wurts & Associates.
Nicholas, J. G. (2000), Market Neutral and Hedged Strategies, in L. Rahl (ed.), Risk
Budgeting: A New Approach to Investing, 2000, Risk Waters Group, London,
pp. 175–248.
Nicholas, J. G. (2002), Marktneutrale Investments – Hedge-Fund-Strategien fur
volatile Markte, 2002, Campus, Frankfurt/New York.
Nye, R. B./Nye, B./Smith, R. C. (1996), Event Investing, in R. A. Lake (ed.), Eval-
uating and Implementing Hedge Fund Strategies – The Experience of Managers
and Investors, 1996, Euromoney Publications, Nestor House, pp. 69–80.
Oberhofer, G. (2001), Hedge Funds – A New Asset Class or Just a Change in
Perspective?, AIMA Newsletter 2001(December).
Okunev, J./White, D. (2002), Smooth Returns and Hedge Fund Risk Factors, 2002,
Working Paper, BT Funds Management and University of New South Wales.
Osterberg, W. P./Thomson, J. B. (1999), The Truth About Hedge Funds, May, 1999,
Federal Reserve Bank of Cleveland.
Paetzmann, K. (2003), Zur Ubertragung von US-Konzepten eines Distressed
Debt Investing auf Deutschland, Zeitschrift fur das gesamte Kreditwesen
56(17), pp. 968–973.
Park, J. M. (1995), Managed Futures as an Investment Asset, 1995, Doctoral Dis-
sertation, Columbia University.
Parnell, R. (2000), To Have and Not Hold, Benefits Canada 24(12), pp. 1–5.
Parnell, R. (2001a), Convertible Bond Arbitrage, Benefits Canada 25(2), pp. 51–55.
Bibliography 285
Parnell, R. (2001b), Hedged Equity – the Long and Short of It, Investments
25, pp. 62–65.
Parnell, R./Matos, R. (2001), Vulture Capitalism, Benefits Canada 25(8), pp. 31–37.
Patel, S. (2003), The S&P Hedge Fund Index – Balancing Representativeness and
Investability, October, 2003, Standard & Poors.
Patel, S. A./Krishnan, B./Meziani, J. (2002), Addressing Risk in Hedge Fund In-
vestments, in B. B. Bruce (ed.), Hedge Fund Strategies – A Global Outlook,
2002, Institutional Investor, New York, pp. 89–97.
Patel, S. A./Roffman, P. A./Meziani, J. (2003), Standard & Poors’s Hedge Fund In-
dex: Structure, Methodology, Definitions and Practices, Journal of Alternative
Investments 6(2), pp. 59–82.
Patterson, A. (2003), Hedge Funds: Hope or Hype, Benefits Canada 27(3), pp. 45–
47.
Patton, A. J. (2002), On the Out-of-Sample Importance of Skewness and Asymmet-
ric Dependence for Asset Allocation, July, 2002, Working Paper, University of
California/London School of Economics.
Paulson, J. (2000), The “Risk” in Risk Arbitrage, in V. Reynolds Parker (ed.),
Managing Hedge Fund Risk: From the Seat of the Practitioner – Views from
Investors, Counterparties, Hedge Funds and Consultants, 2000, Risk Books,
London, pp. 189–199.
Peetz, D./Compton, P. (2003), Wandelanleihenarbitrage: Attraktiv aber nicht ohne
Risiken, Die Bank 2003(3), pp. 202–206.
Perli, R./Sack, B. (2003), Does Mortgage Hedgeing Amplify Movements in Long-
Term Interest Rates?, Journal of Fixed Income 13(3), pp. 7–17.
Petersmeier, K. (2003), Kerndichte- und Kernregressionsschatzungen im Asset Man-
agement, 2003, Uhlenbruch, Bad Soden/Ts.
Bibliography 286
Press, J./Stamatelos, T. (1996), Assessing Risk and Risk Control – Operational
Issues, in R. A. Lake (ed.), Evaluating and Implementing Hedge Fund Strategies
– The Experience of Managers and Investors, 1996, Euromoney Publications,
Nestor House, pp. 201–216.
PriceWaterhouseCoopers (2003), The Regulation and Distribution of Hedge Funds
in Europe, 2003, PriceWaterhouseCoopers.
Perisse, G. (2004), Hedgefonds – Strategien, Risiken und Chancen fur den privaten
Investor, 2004, Societe Generale Asset Management.
Rahl, L. (2000), Risk Budgeting: The Next Stop of the Risk Managment Journey –
The Veteran´s Perspective, in L. Rahl (ed.), Risk Budgeting: A New Approach
to Investing, 2000, Risk Waters Group, London, pp. 3–26.
Rehkugler, H./Jandura, D. (2002), Kointegrations- und Fehlerkorrekturmodelle zur
Finanzmarktprognose, in J. M. Kleeberg/H. Rehkugler (eds), Handbuch Port-
foliomanagement, 2002, Uhlenbruch, Bad Soden/Ts., pp. 649–685.
Reinganum, M. R. (1981), Misspecifications of Capital Asset Pricing – Empirical
Anomalies Based on Earnings and Market Values, Journal of Financial Eco-
nomics 9(1), pp. 19–46.
Reiss, R. (1996), Value Investing, in R. A. Lake (ed.), Evaluating and Implement-
ing Hedge Fund Strategies – The Experience of Managers and Investors, 1996,
Euromoney Publications, Nestor House, pp. 49–54.
Reserve Bank of Australia (1999), Hedge Funds, Financial Stability and Market
Integrity, June, 1999.
Rettberg, U. (2003), Aufsichtsbehorden wollen Hedge Fonds starker kontrollieren,
Handelsblatt 2003(March 27), p. 37.
Reynolds Parker, V./Warsager, R. G. (2000), The Diversity and Commonality of
Risk, in V. Reynolds Parker (ed.), Managing Hedge Fund Risk: From the Seat
Bibliography 287
of the Practitioner – Views from Investors, Counterparties, Hedge Funds and
Consultants, 2000, Risk Books, London, pp. XXI–XXXXVII.
Ringoen, G. W. (1996), Short Selling as an Alternative Investment Strategy, in
R. A. Lake (ed.), Evaluating and Implementing Hedge Fund Strategies – The
Experience of Managers and Investors, 1996, Euromoney Publications, Nestor
House, pp. 89–100.
Romano, C. (2002a), Applying Copula Function to Risk Management, 2002, Working
Paper, Banca di Roma.
Romano, C. (2002b), Calibrating and Simulating Copula Functions: An Application
to the Italian Stock Market, 2002, Working Paper, Centro Interdipartimentale
sul Diritto e l’Economia dei Mercati.
Rosenberg, B./Reid, K./Lanstein, R. (1985), Persuasive Evidence of Market Ineffi-
ciency, Journal of Portfolio Management 11(3), pp. 9–17.
Roy, A. D. (1952), Safety-First and the Holding of Assets, Econometrica
20(3), pp. 431–449.
Rubinstein, M. E. (1973), The Fundamental Theorem of Parameter-Preference Secu-
rity Valuation, Journal of Financial and Quantitative Analysis 8(1), pp. 61–69.
Rubinstein, M. E. (1976), The Valuation of Uncertain Income Streams and the
Pricing of Options, Bell Journal of Economics and Management Science
7(2), pp. 407–425.
Ruckstuhl, T./Meier, P./Lodeiro, S./Kundig, O. (2004), Funds of Hedge Funds In-
dices: Properties, Purposes and Representativeness, 2004, Working Paper, In-
stitut Banking & Finance Zuricher Hochschule Winterthur.
Rutkis, A. (2002), Hedge-Fonds als alternative Investments, 2002, Bankakademie,
Frankfurt.
Bibliography 288
Saerfvenblad, P. (2002), Creating Returns and Contolling Risk in Managed Funds,
2002, RPM Risk & Portfolio Management AB.
Safaty, N. (2001), The $460bn Global Convertible Bond Market, Fixed Income Mar-
ket Review 2001(July), pp. 109–116.
Samuelson, P. A. (1967), General Proof That Diversification Pays, Journal of Fi-
nancial and Quantitative Analysis 2(2), pp. 1–13.
Samuelson, P. A. (1970), The Fundamental Approximation Theorem of Portfolio
Analysis in Terms of Means, Variances and Higher Moments, Review of Eco-
nomic Studies 37(4), pp. 537–542.
Scherer, B. (2002), Portfolio Construction and Risk Budgeting, 2002, Risk Waters,
London.
Schiendl, P./Hofer, B. (2003), Minimum Requirements Concerning Reporting for
Fund of Funds, in HVB Alternatives (ed.), Activity Report 2003, 2003, pp. 40–
42.
Schlittgen, R./Streitberg, B. (1999), Zeitreihenanalyse, 8th edn, 1999, Oldenbourg,
Munchen/Wien.
Schmidt, J. H. (2003), The Return of Global Macro: Why Investors Should Be Wary
of Irrational Expectations, SwissHEDGE 3(1), pp. 23–27.
Schmittmann, P./Schaffeler, U. (2002), Kompensationszahlungen bei der Aktien-
leihe nach der Steuerreform, FinanzBetrieb 4(9), pp. 530–537.
Schneeweis, T./Kazemi, H./Martin, G. (2001), Understanding Hedge Fund Perfor-
mance: Research Results and Rules of Thumb for the Institutional Investor,
November, 2001, University of Massachussets/Lehman Brothers.
Schneeweis, T./Kazemi, H./Martin, G. (2002), Understanding Hedge Fund Perfor-
mance: Research Issues Revisited – Part 1, Journal of Alternative Investments
5(3), pp. 4–22.
Bibliography 289
Schneeweis, T./Spurgin, R. (2000), Hedge Funds: Portfolio Risk Diversifiers, Return
Enhancers or Both?, July, 2000, Working Paper, University of Massachusetts.
Schubert, L. (2005), Performance of Linear Portfolio Optimization, 2005, Working
Paper, Constance University of Applied Sciences.
Scott, D. W. (1992), Multivariate Density Estimation: Theory, Practice, and Visu-
alization, 1992, John Wiley & Sons, New York et al.
Scott, R. C./Horvath, P. A. (1980), On the Direction of Preference for Moments of
Higher Order Than the Variance, Journal of Finance 35(4), pp. 915–919.
Seides, T. (2002), A Matter of Trust: The Issue of Risk Transparency, in B. B. Bruce
(ed.), Hedge Fund Strategies – A Global Outlook, 2002, Institutional Investor,
New York, pp. 103–110.
Signer, A. (2002), Verzerrungen in Hedge Fund-Renditen und Ihre Ursachen, Absolut
Report 6(06/2002), pp. 28–35.
Signer, A. (2003), Generieren Hedge Funds einen Mehrwert?, 2003, Paul Haupt,
Bern.
Silverman, B. W. (1986), Density Estimation for Statistics and Data Analysis, 1986,
Chapman & Hall/CRC, Boca Raton et al.
Simkowitz, M. A./Beedles, W. L. (1978), Diversification in a Three-Moment World,
Journal of Financial and Quantitative Analysis 13(5), pp. 927–941.
Sklar, A. (1959), Fonctions de repartion a n dimensions et leurs marges, Publ. Inst.
Statist. Univ. Paris 8, pp. 229–231.
Smith, M. D. (2001), Pitfalls of Hedge Funds, 2001, Southern Employees Benefits
Conference, Hewitt Investment Group.
Speidell, L. S. (2002), What is Emerging in Emerging Markets?, Journal of Investing
11(2), pp. 5–14.
Bibliography 290
Spitz, W. T. (2001), Selecting Hedge Funds: The Vanderbilt Perspective, in AIMR
(ed.), Hedge Fund Management, 2001, AIMR Conference Proceedings, pp. 60–
61.
Standard & Poor’s (2003a), Standard & Poor’s Hedge Fund Index Series: Frequently
Asked Questions, May, 2003.
Standard & Poor’s (2003b), Standard & Poor’s Hedge Fund Index: Structure,
Methodology, Definitions, and Practices, January, 2003.
Standard & Poor’s (2004), Standard & Poor’s Equity Long/Short Index: Structure,
Methodology, Definitions, and Practices, May, 2004.
State Street (2005), Hedge Fund Research Study: Key Findings, 2005.
Steiner, M./Bruns, C. (2002), Wertpapiermanagement, 8th edn, 2002, Schaffer-
Poeschel, Stuttgart.
Stemme, K./Slattery, P. (2002), Hedge Fund Investments: Do It Yourself or Hire a
Contractor, in B. B. Bruce (ed.), Hedge Fund Strategies – A Global Outlook,
2002, Institutional Investor, New York, pp. 60–68.
Stonham, P. (1999), Too Close to the Hedge: The Case of Long Term Capital
Management LP – Part One: Hedge Fund Analytics, European Management
Journal 17(3), pp. 282–289.
Strome, M. (1996), Macro Investing, in R. A. Lake (ed.), Evaluating and Imple-
menting Hedge Fund Strategies – The Experience of Managers and Investors,
1996, Euromoney Publications, Nestor House, pp. 27–34.
Supreme Court of The State of New York (2003), Complaint Against Canary Cap-
ital Partners, LLC, Canary Investment Management, LLC, Canary Capital
Partners, LTD and Edward J. Stern, 2003, September 3.
Tarrant, J. (1996), Hedge Fund Investing: A Private Family Perspective, Evaluating
Bibliography 291
and Implementing Hedge Fund Strategies – The Experience of Managers and
Investors, 1996, Euromoney Publications, Nestor House, pp. 145–152.
Thode, H. C. (2002), Testing for Normality, 2002, Marcel Dekker, New York.
Till, H. (2003), Weighting the Cost of Illiquidity, RISK 16(11), pp. S13–S16.
Tisdale, J. (2000), Risk Management for Hedge Fund Strategies: US Equity Market
Neutral, in V. Reynolds Parker (ed.), Managing Hedge Fund Risk: From the
Seat of the Practitioner – Views from Investors, Counterparties, Hedge Funds
and Consultants, 2000, Risk Books, London, pp. 169–175.
Tobin, J. (1958), Liquidity Preference as Behavior Towards Risk, Review of Eco-
nomic Studies 25(67), pp. 68–85.
Tobin, J. (1969), Comment on Borch and Feldstein, Review of Economic Studies
36(105), pp. 13–14.
Topaloglou, N./Vladimirou, H./Zenios, S. A. (2002), CVaR Models with Selective
Hedging for International Asset Allocation, Journal of Banking and Finance
26(7), pp. 1535–1561.
Tremont (1999), Mortgage-Backed Securities in the Hedge Fund Marketplace,
November, 1999, Tremont Advisers Inc.
Tremont (2000a), Convertible Arbitrage: Opportunity & Risk, September, 2000,
Tremont Advisers Inc.
Tremont (2000b), Distressed Securities Investing, November, 2000, Tremont Advis-
ers Inc.
Tsiang, S. (1972), The Rationale of the Mean-Standard Deviation Analysis, Skew-
ness Preference, and the Demand for Money, American Economic Review
62(3), pp. 354–371.
Tuckman, B. (2002), Fixed Income Securities: Tools for Today‘s Markets, 2002,
John Wiley & Sons, New York et al.
Bibliography 292
UBS (2004), European Rainmakers, April, 2004, UBS Investment Research.
Uryasev, S. (2000), Introduction to the Theory of Probabilistic Constrained Opti-
mization (Value-at-Risk), in S. Uryasev (ed.), Probabilistic Constrained Opti-
mization: Methodology and Applications, 2000, Kluwer Academic Publishers,
pp. 1–25.
Van, G. P. (1996), Quantitative Analysis of Hedge Fund Return/Risk Characteris-
tics, in R. A. Lake (ed.), Evaluating and Implementing Hedge Fund Strategies
– The Experience of Managers and Investors, 1996, Euromoney Publications,
Nestor House, pp. 217–252.
Wallmeier, M. (1997), Prognose von Aktienrenditen und -risiken mit Mehrfaktoren-
modellen, 1997, Uhlenbruch, Bad Soden/Ts.
Watson Wyatt (2005), Capacity in the Hedge Fund Industry, March, 2005, Watson
Wyatt LLP Investment Consulting.
Wilmot-Smith, J. (2003), Requirements to Fund of Hedge Funds from an Investor’s
Point of View and Their Meaning in Strategic Asset Allocation, in HVB Alter-
natives (ed.), Activity Report 2003, 2003, pp. 65–67.
Wilmott, P. (2000a), Paul Wilmott on Quantitative Finance: Volume One, 2000,
John Wiley & Sons, New York et al.
Wilmott, P. (2000b), Paul Wilmott on Quantitative Finance: Volume Two, 2000,
John Wiley & Sons, New York et al.
Wintner, B. A. (2001), How Many Hedge Funds Are Needes to Create a Diversified
Fund or Funds?, March, 2001, Asset Alliance.
Wong, A. M. (1991), Trading and Investing in Bond Options, 1991, John Wiley &
Sons, New York et al.
Wong, A. M. (1993), Fixed-Income Arbitrage, 1993, John Wiley & Sons, New York
et al.
Bibliography 293
Wong, W.-K./Au, T. K.-K. (2002), On Two-Moment Decision Models and Ex-
pected Utility Maximization, 2002, Working Paper, National University of Sin-
gapore/The Chinese University of Hong Kong.
World Bank (2004), World Development Indicators 2004, 2004.
Wyser-Pratte, G. P. (1982), Risk Arbitrage II, Monograph Series in Finance and
Economics, 1982, Salomon Brothers Center for the Study of Financial Institu-
tions/Graduate School of Business Administration New York University.
Yamai, Y./Yoshiba, T. (2005), Value-at-Risk versus Expected Shortfall: A Practical
Perspective, Journal of Banking and Finance 29(4), pp. 997–1015.
Yang, P. L./Faux, R. G. (1996), Managed Futures – The Convergence with Hedge
Funds, in R. A. Lake (ed.), Evaluating and Implementing Hedge Fund Strategies
– The Experience of Managers and Investors, 1996, Euromoney Publications,
Nestor House, pp. 81–88.
Yang, T./Branch, B. (2001), Merger Arbitrage: Evidence of Profitability, Journal
of Alternative Investments 4(3), pp. 17–32.
Yee, M. (2004), Convertible Arbitrage Hedge-Fund-Management, Absolut Report
18(02/2004), pp. 20–25.
Young, M. R. (1998), A Minimax Portfolio Selection Rule with Linear Programming
Solution, Management Science 44(5), pp. 673–683.
Zagst, R. (2002), Interest Rate Management, 2002, Springer, New York et al.
Zask, E./Bousbib, G./Ewing, P. (2004), Can the Hedge Fund Industry Mature With-
out Investable Indices?, AIMA Newsletter 2004(September).
Zitzewitz, E. (2003), Who Cares About Shareholders? Arbitrage-Proofing Mutual
Funds, Journal of Law, Economics, and Organization 19(2), pp. 245–280.
Zivot, E./Wang, J. (2003), Modeling Financial Time Series with S-Plus, 2003, In-
sightful.